ProgrammableMarketing/inferact.ai (opens in a new tab)
PDF

Views

  • Plan
  • Roadmap
  • Research

On this page

  • The brief
  • Start here
  • Idea portfolio
  • Effort vs impact
  • This week · 6
  • This quarter · 10
  • Big bets · 3
  • Wildcards · 2
  • How we can help
Generated Oct 23, 2026
  1. Programmable
  2. Playbooks
  3. Inferact
Inferact logo

growth playbook · Oct 23, 2026

inferact.ai (opens in a new tab)

Inferact growth playbook

Diagnosis

Inferact has rare assets: the team behind vLLM, a $150 million seed, Google Cloud and AMD partnerships, day-0 model support and headline benchmark results.

It is now hiring its first marketing and GTM team: a Technical PMM for launches and the benchmark program, a social manager, a designer who owns every preview image and the Webflow site, a GTM engineer to build the CRM from scratch, an events PMM and a partnerships manager.

The site has not caught up: 10 pages, no product, pricing or case-study pages, homepage numbers without sources, a Typeform for meetings, and a newest launch post that shares without a preview image and is missing from the sitemap. This playbook fixes the launch basics first, then builds product pages, a defined assessment offer, a benchmark dashboard and community-sourced pipeline.

Growth Score

7/10

Biggest Opportunity

Turn every launch and every vLLM user into measurable pipeline: shareable, sourced launch posts, product pages and a concrete assessment offer, routed into a CRM from day one.

Biggest Risk

Baseten, Fireworks, Together AI and the SGLang company sell to the same teams, and open-source users can stay on free vLLM if the paid platform's value is not clear and proven.

01

Start here

The three moves to make first. Each one is written up in full below.

  1. 1

    Make every launch post shareable and findable: the newest benchmark post has no preview image and is not in the sitemap

    Upload a share image for the AgentX post and republish the sitemap.

    High impact First result in 1 week
  2. 2

    Link every homepage number to its source

    Collect the source for each of the four numbers.

    High impact First result in 1 week
  3. 3

    Give Inferact Platform and Inferact Cloud their own pages

    Outline both pages on one sheet.

    High impact First result in 1 month

02

Your idea portfolio

21 ideas, grouped by when to run them. Open any idea for how to run it: the steps, who to approach, copy you can use, and how to measure it.

  • 6 ideas are marked Unexpected: plays you are unlikely to have tried.

Every idea placed by the effort it takes against the impact it could have. Quick wins sit top left; select a number to open that idea. The filters below narrow the list by status, timing, category and what Programmable can build.

Effort vs impactEach number is an idea. Select one to open it.
Impact →
Quick wins · 3Big bets · 13Fill-ins · 3Reconsider · 2
Effort →

Showing 21 of 21 ideas

This week 6

Quick wins with a clear first action.

  • 01

    Inferact grows through launches: benchmark posts, partner announcements and day-0 model support. Its newest post, the TPU megakernel results on SemiAnalysis's AgentX benchmark, is featured on the homepage and news page, but its share image fields are empty and it is missing from the sitemap. Every link to it on X, LinkedIn, Hacker News or Slack shows no preview card, and search engines are not told it exists. The company is hiring a Technical PMM, a Social Media Manager and a designer precisely to own "launch communications", "partner amplification" and "every... social card, and preview image that ships".

    High impactLow effort First result in 1 week Technical Product Marketing M…

    First action

    Upload a share image for the AgentX post and republish the sitemap.

    How to measure it

    Launch-post sessions in week one

    Now
    –
    Target
    +30%
    By
    90 days

    Steps

    1. 1

      Fix the post

      Share image and twitter:image on the AgentX post.

      Graphic Designer 1 hour

    2. 2

      Republish sitemap

      Confirm all six posts are listed.

      Graphic Designer 1 hour

    3. 3

      Template rule

      Required share image with branded fallback.

      Graphic Designer 1 day

    4. 4

      Launch checklist

      Card, sitemap, sources, partner copy, UTMs.

      Technical Product Marketing Manager 2 days

    5. 5

      Preview test

      Check cards on X, LinkedIn and Slack before posting.

      Social Media Manager each launch

    6. 6

      Partner kit

      Same card and copy sent to partners.

      Technical Product Marketing Manager each launch

    7. 7

      Measure

      Sessions and sign-ups per launch.

      GTM Engineer monthly

    First result
    1 week
    Cost
    Staff time only
    Owner
    Technical Product Marketing Manager

    Smallest test

    Fix the one post and compare its shares with the previous launch.

    Why it works

    Fix the post today (share image, sitemap). Then make it impossible to repeat: in the Webflow blog template, make the share image a required field with a branded fallback, confirm the sitemap updates on publish, and add a launch checklist (share card, sitemap, source links, partner copy, UTM links) that the TPMM, social and design hires sign off before every post goes live. Measure clicks per launch from social and search.

    A preview card typically lifts click-through on shared links substantially, and launches are Inferact's main way to reach engineers. Getting the basics right on every post multiplies the reach of work the engineers already did.

    Ready to use

    ChecklistLaunch post checklist

    [ ] Share image 1200x630 set (og:image and twitter:image)
    [ ] Post appears in sitemap.xml
    [ ] Every number links to its source
    [ ] Title and description written for search
    [ ] Partner copy and card sent
    [ ] UTM links for each channel
    [ ] Preview checked on X, LinkedIn and Slack
    [ ] Week-one report

    Risks and how to handle them

    • Checklist slows launches

      Keep it to one screen and automate the checks.

    • Webflow sitemap settings

      Check the post is not excluded at collection level.

    Evidence

    • Your websiteThe newest post, "Serving agentic workloads with TPU megakernels in vLLM" (October 2026, featured on the news page and the homepage), has an empty og:image and twitter:image, so links shared on X, LinkedIn or Slack show no preview image, and it is missing from the sitemap. Every earlier post has its own share image. Checked twice on October 22, 2026.Source
    • HiringTechnical Product Marketing Manager (posted September 24, 2026): owns "day-0 launch communications, product positioning, the benchmark narrative, customer proof"; within 60 days owns "website product pages" and "partnership announcements with ecosystem partners such as Google, AMD, and NVIDIA"; within 90 days presents a plan covering "launch tiers, the benchmark program, case study pipeline", and builds a launch process covering "intake, messaging, review, distribution, partner amplification, and post-launch measurement".Source
    • HiringSocial Media Manager (posted September 24, 2026): owns "the daily voice, publishing system, and audience growth strategy for vLLM and Inferact"; "Own the social side of the rebrand launch alongside the website"; a voice guide separating "the vLLM community voice from the Inferact company voice"; a founder program; testing "Hacker News, Instagram, and TikTok"; a "partner amplification network"; and a "repeatable social launch system with tiered templates, asset checklists".Source
    • HiringGraphic Designer (posted September 24, 2026): "Own every blog diagram, partnership announcement kit, social card, and preview image that ships"; build Webflow "event pages, partnership announcement pages, blog templates"; later own "product pages, the public benchmark dashboard home, and conference sites for vLLM Conference US and China".Source
  • 02

    The homepage leads with four proof points (464 tok/s, 709 tok/s/user, 18.3M tokens per dollar, #1 on Artificial Analysis), but only one links to where it came from. Infrastructure engineers check numbers before they trust them.

    High impactLow effort First result in 1 week Technical Product Marketing M…

    First action

    Collect the source for each of the four numbers.

    How to measure it

    Proof-strip clicks

    Now
    –
    Target
    5% of visits
    By
    60 days

    Steps

    1. 1

      Find sources

      Post, run or leaderboard for each.

      Technical Product Marketing Manager 1 day

    2. 2

      Add links

      Each number to its source.

      Graphic Designer 2 hours

    3. 3

      Methods note

      Date, hardware, configuration.

      Technical Product Marketing Manager 1 day

    4. 4

      Refresh rule

      Update when a number is beaten.

      Technical Product Marketing Manager monthly

    5. 5

      Track

      Clicks per number.

      GTM Engineer monthly

    6. 6

      Reuse

      Same sources in decks.

      Technical Product Marketing Manager ongoing

    First result
    1 week
    Cost
    $0
    Owner
    Technical Product Marketing Manager

    Smallest test

    Link the 18.3M figure to the AgentX post today.

    Why it works

    Link each figure to its post, benchmark run or leaderboard snapshot, with the date and configuration. Add a short methods note under the strip.

    Sourced numbers convert skeptical engineers; unsourced ones get discounted, and the benchmark narrative is the TPMM's to own.

    Risks and how to handle them

    • Numbers age quickly

      Date each figure.

    Evidence

    • Your websiteHomepage proof numbers: "464 tok/s bs=1 decode on Kimi-K3 on launch day", "709 tok/s/user with first Megakernel on TPU", "18.3M tok / 1$ On Semianalysis AgentX", "#1 On Artificial Analysis, May 2026". Only the 709 tok/s figure links to its source post.Source
    • Your websiteThe newest post, "Serving agentic workloads with TPU megakernels in vLLM" (October 2026, featured on the news page and the homepage), has an empty og:image and twitter:image, so links shared on X, LinkedIn or Slack show no preview image, and it is missing from the sitemap. Every earlier post has its own share image. Checked twice on October 22, 2026.Source
    • HiringTechnical Product Marketing Manager (posted September 24, 2026): owns "day-0 launch communications, product positioning, the benchmark narrative, customer proof"; within 60 days owns "website product pages" and "partnership announcements with ecosystem partners such as Google, AMD, and NVIDIA"; within 90 days presents a plan covering "launch tiers, the benchmark program, case study pipeline", and builds a launch process covering "intake, messaging, review, distribution, partner amplification, and post-launch measurement".Source
  • 03

    The site has 10 pages. The two products, on your compute or on Inferact's, exist only as homepage sections, and /platform, /cloud and /pricing return 404. Buyers searching for a self-hosted vLLM platform or managed vLLM inference have nowhere specific to land.

    High impactMedium effort First result in 1 month Technical Product Marketing M…

    First action

    Outline both pages on one sheet.

    How to measure it

    Product-page meetings

    Now
    –
    Target
    20 a month
    By
    90 days

    Steps

    1. 1

      Outline

      Audience, model, features.

      Technical Product Marketing Manager 3 days

    2. 2

      Design

      Two pages in Webflow.

      Graphic Designer 2 weeks

    3. 3

      Proof

      Quote and sourced numbers.

      Technical Product Marketing Manager 1 week

    4. 4

      Forms

      Into the CRM.

      GTM Engineer 2 days

    5. 5

      Publish

      With search titles.

      Graphic Designer 1 day

    6. 6

      Measure

      Meetings per page.

      GTM Engineer monthly

    First result
    1 month
    Cost
    Staff time only
    Owner
    Technical Product Marketing Manager

    Smallest test

    Publish the Platform page first.

    Why it works

    Build one page per product: who it is for, deployment model, what is included beyond open-source vLLM, security, a quote, sourced benchmarks and a "talk to an engineer" form. Add a comparison table between the two.

    Product pages capture search and give sales one link per buying situation.

    Risks and how to handle them

    • Product names change in the rebrand

      Build pages on the new names.

    Evidence

    • Your websiteThe sitemap has 10 URLs: home, careers, news, privacy, terms and five news posts. Inferact Platform ("Run inference on your compute") and Inferact Cloud ("Run inference on our compute") exist only as homepage sections; /platform, /cloud, /pricing and /benchmarks return 404. There is no documentation, pricing or case-study page.Source
    • Your websiteHomepage: "From the creators of vLLM. Frontier inference you control."; features include day-zero models, autoscaling, "Frontier Tuning", endpoints and metering, observability, "Forward-deployed Inference Engineers" and SOC 2; "vLLM is optimized for everyone. Inferact is optimized for you." The main call to action, "Book a meeting", opens a Typeform; the footer says "Tell us a model to serve. We'll help optimize it. Talk to an engineer".Source
    • HiringTechnical Product Marketing Manager (posted September 24, 2026): owns "day-0 launch communications, product positioning, the benchmark narrative, customer proof"; within 60 days owns "website product pages" and "partnership announcements with ecosystem partners such as Google, AMD, and NVIDIA"; within 90 days presents a plan covering "launch tiers, the benchmark program, case study pipeline", and builds a launch process covering "intake, messaging, review, distribution, partner amplification, and post-launch measurement".Source
    • HiringGraphic Designer (posted September 24, 2026): "Own every blog diagram, partnership announcement kit, social card, and preview image that ships"; build Webflow "event pages, partnership announcement pages, blog templates"; later own "product pages, the public benchmark dashboard home, and conference sites for vLLM Conference US and China".Source
  • 04

    The main call to action opens a Typeform, and the GTM Engineer is hired to build "a clean commercial source of truth" from scratch. Inbound requests should arrive in that system already sorted.

    High impactLow effort First result in 2 weeks GTM Engineer

    First action

    Draft five qualifying questions.

    How to measure it

    Request-to-call rate

    Now
    –
    Target
    60%
    By
    60 days

    Steps

    1. 1

      Questions

      Five fields.

      GTM Engineer 1 day

    2. 2

      Build

      Form into CRM.

      GTM Engineer 3 days

    3. 3

      Routing

      High-fit to FDEs.

      GTM Engineer 2 days

    4. 4

      Replace links

      Header and footer.

      Graphic Designer 1 hour

    5. 5

      Reply SLA

      One business day.

      Head of Commercial ongoing

    6. 6

      Report

      Weekly.

      GTM Engineer weekly

    First result
    2 weeks
    Cost
    $0
    Owner
    GTM Engineer

    Smallest test

    Swap the header link to the new form.

    Why it works

    Use a form on the site that asks model, hardware, monthly tokens or GPUs, and own-compute versus Inferact Cloud, writes straight to the CRM, and routes high-fit requests to a forward-deployed engineer within a day.

    Qualified, routed inbound is the cheapest pipeline a young sales team can get.

    Risks and how to handle them

    • Longer form, fewer requests

      Keep five fields; enrich the rest.

    Evidence

    • Your websiteHomepage: "From the creators of vLLM. Frontier inference you control."; features include day-zero models, autoscaling, "Frontier Tuning", endpoints and metering, observability, "Forward-deployed Inference Engineers" and SOC 2; "vLLM is optimized for everyone. Inferact is optimized for you." The main call to action, "Book a meeting", opens a Typeform; the footer says "Tell us a model to serve. We'll help optimize it. Talk to an engineer".Source
    • HiringGTM Engineer (posted September 29, 2026): "build the commercial operating system for Inferact from the ground up", where "CRM structure, account intelligence, standardized workflows" are "still being defined"; early projects include "a clean commercial source of truth" and "mining Inferact's unique ecosystem and community data to identify high-value prospects and warm paths".Source
    • HiringDeveloper Relations Engineer: "make vLLM the default way developers understand, build, and scale AI inference" through deep dives, demos, tutorials and workshops on "KV cache, continuous batching, prefix caching, prefill and decode, quantization, GPU serving, latency versus throughput". Forward Deployed Engineer (posted October 2, 2026): deploys vLLM-powered inference in customer environments, including "enterprise, regulated, security-sensitive, or bring-your-own-cloud environments", and affects "customer time-to-value".Source
  • 05

    The footer invites visitors to "Tell us a model to serve. We'll help optimize it." That is a strong offer with no definition: what is measured, how long it takes, what the buyer gets.

    High impactMedium effort First result in 1 month Forward Deployed Engineer

    First action

    Write the report template.

    How to measure it

    Assessments started

    Now
    –
    Target
    10 a month
    By
    90 days

    Steps

    1. 1

      Define scope

      Model, hardware, metrics.

      Forward Deployed Engineer 1 week

    2. 2

      Template

      Report format.

      Technical Product Marketing Manager 1 week

    3. 3

      Page

      Offer landing page.

      Graphic Designer 1 week

    4. 4

      Intake

      Form fields.

      GTM Engineer 2 days

    5. 5

      Run

      Five-day delivery.

      Forward Deployed Engineer ongoing

    6. 6

      Track

      Assessment to contract.

      GTM Engineer monthly

    First result
    1 month
    Cost
    Engineering time
    Owner
    Forward Deployed Engineer

    Smallest test

    Run two assessments by hand.

    Why it works

    Package it: one model, the buyer's hardware or Inferact's, a written report on tokens per dollar and latency against their current setup in five working days. Put it on the product pages and in outbound.

    A concrete, measurable offer gets more requests than "book a meeting", and the report becomes the business case.

    Risks and how to handle them

    • FDE time overrun

      Cap assessments per month.

    Evidence

    • Your websiteHomepage: "From the creators of vLLM. Frontier inference you control."; features include day-zero models, autoscaling, "Frontier Tuning", endpoints and metering, observability, "Forward-deployed Inference Engineers" and SOC 2; "vLLM is optimized for everyone. Inferact is optimized for you." The main call to action, "Book a meeting", opens a Typeform; the footer says "Tell us a model to serve. We'll help optimize it. Talk to an engineer".Source
    • HiringDeveloper Relations Engineer: "make vLLM the default way developers understand, build, and scale AI inference" through deep dives, demos, tutorials and workshops on "KV cache, continuous batching, prefix caching, prefill and decode, quantization, GPU serving, latency versus throughput". Forward Deployed Engineer (posted October 2, 2026): deploys vLLM-powered inference in customer environments, including "enterprise, regulated, security-sensitive, or bring-your-own-cloud environments", and affects "customer time-to-value".Source
    • Your websiteHomepage proof numbers: "464 tok/s bs=1 decode on Kimi-K3 on launch day", "709 tok/s/user with first Megakernel on TPU", "18.3M tok / 1$ On Semianalysis AgentX", "#1 On Artificial Analysis, May 2026". Only the 709 tok/s figure links to its source post.Source
  • 06

    Inferact says "vLLM is optimized for everyone. Inferact is optimized for you", and the Social Media Manager will write a guide separating the two voices. Engineers who love vLLM ask the same questions: what stays open, who governs it, what costs money.

    Medium impactLow effort First result in 2 weeks Social Media Manager

    First action

    Collect the ten most common questions from sales calls and social replies.

    How to measure it

    FAQ visits

    Now
    –
    Target
    2,000 a month
    By
    90 days

    Steps

    1. 1

      Collect questions

      Calls and replies.

      Social Media Manager 1 week

    2. 2

      Draft

      Plain answers.

      Technical Product Marketing Manager 3 days

    3. 3

      Founder review

      Governance wording.

      Co-Founder and CEO 2 days

    4. 4

      Publish

      With FAQ markup.

      Graphic Designer 1 day

    5. 5

      Link

      Bios and homepage.

      Social Media Manager 1 day

    6. 6

      Update

      Each release.

      Technical Product Marketing Manager quarterly

    First result
    2 weeks
    Cost
    $0
    Owner
    Social Media Manager

    Smallest test

    Post the FAQ as a thread first.

    Why it works

    Publish a plain FAQ: vLLM stays Apache 2.0 under the PyTorch Foundation; what the platform adds (control plane, tuning, day-0 models, metering, engineers); when teams need it. Link it from social bios and the homepage.

    Clear answers protect community trust and remove the main objection in sales.

    Risks and how to handle them

    • Community reads it as a sales page

      Keep it factual and link to the open project first.

    Evidence

    • Your websiteHomepage: "From the creators of vLLM. Frontier inference you control."; features include day-zero models, autoscaling, "Frontier Tuning", endpoints and metering, observability, "Forward-deployed Inference Engineers" and SOC 2; "vLLM is optimized for everyone. Inferact is optimized for you." The main call to action, "Book a meeting", opens a Typeform; the footer says "Tell us a model to serve. We'll help optimize it. Talk to an engineer".Source
    • Your websiteNews: Google Cloud and Inferact announced "a joint engineering and go-to-market partnership to make TPU a first-class citizen in vLLM" (September 14, 2026); vLLM "supports more than 5,000 model architectures, counts a community of over 3,000 contributors"; an AMD collaboration on AMD Instinct GPUs (October 2026); full Kimi K3 vendor verification (August 2026). vLLM is "an Apache 2.0 project governed by the PyTorch foundation".Source
    • HiringSocial Media Manager (posted September 24, 2026): owns "the daily voice, publishing system, and audience growth strategy for vLLM and Inferact"; "Own the social side of the rebrand launch alongside the website"; a voice guide separating "the vLLM community voice from the Inferact company voice"; a founder program; testing "Hacker News, Instagram, and TikTok"; a "partner amplification network"; and a "repeatable social launch system with tiered templates, asset checklists".Source

This quarter 10

The core plays for the next three months.

  • 07

    The designer's 90-day plan includes "the public benchmark dashboard home", and the TPMM owns "the benchmark program". Today benchmark numbers live in individual posts.

    High impactHigh effort First result in 1 quarter Technical Product Marketing M…

    First action

    Pick the five model and hardware pairs to publish first.

    How to measure it

    Dashboard-sourced meetings

    Now
    –
    Target
    15 a month
    By
    2 quarters

    Steps

    1. 1

      Pick pairs

      Five to start.

      Technical Product Marketing Manager 1 week

    2. 2

      Configs

      Public and reproducible.

      Performance team 1 month

    3. 3

      Build

      Dashboard in Webflow or a light app.

      Graphic Designer 1 month

    4. 4

      Automate

      Refresh per release.

      Performance team 1 month

    5. 5

      Launch

      With the launch checklist.

      Social Media Manager 1 week

    6. 6

      Measure

      Visits and meetings.

      GTM Engineer monthly

    First result
    1 quarter
    Cost
    Engineering time
    Owner
    Technical Product Marketing Manager

    Smallest test

    Publish one model and hardware pair.

    Why it works

    Build one dashboard with tokens per dollar, latency and throughput by model and hardware, each linked to its configuration and run date, and refresh it with each release.

    Reproducible numbers are what infrastructure buyers trust, and a living dashboard earns links and AI citations every week.

    Risks and how to handle them

    • Rivals dispute numbers

      Publish configs and invite reruns.

    Evidence

    • HiringTechnical Product Marketing Manager (posted September 24, 2026): owns "day-0 launch communications, product positioning, the benchmark narrative, customer proof"; within 60 days owns "website product pages" and "partnership announcements with ecosystem partners such as Google, AMD, and NVIDIA"; within 90 days presents a plan covering "launch tiers, the benchmark program, case study pipeline", and builds a launch process covering "intake, messaging, review, distribution, partner amplification, and post-launch measurement".Source
    • HiringGraphic Designer (posted September 24, 2026): "Own every blog diagram, partnership announcement kit, social card, and preview image that ships"; build Webflow "event pages, partnership announcement pages, blog templates"; later own "product pages, the public benchmark dashboard home, and conference sites for vLLM Conference US and China".Source
    • Your websiteHomepage proof numbers: "464 tok/s bs=1 decode on Kimi-K3 on launch day", "709 tok/s/user with first Megakernel on TPU", "18.3M tok / 1$ On Semianalysis AgentX", "#1 On Artificial Analysis, May 2026". Only the 709 tok/s figure links to its source post.Source
    • Your websiteThe newest post, "Serving agentic workloads with TPU megakernels in vLLM" (October 2026, featured on the news page and the homepage), has an empty og:image and twitter:image, so links shared on X, LinkedIn or Slack show no preview image, and it is missing from the sitemap. Every earlier post has its own share image. Checked twice on October 22, 2026.Source
  • 08

    Day-0 support is a headline promise ("Frontier models served under your control the day they launch"), and the Kimi K3 work produced a launch-day number. Each big model release is a burst of searches from teams who want to serve it.

    High impactMedium effort First result in 1 quarter Technical Product Marketing M…

    First action

    Build the template using the Kimi K3 launch.

    How to measure it

    Day-0 meetings

    Now
    –
    Target
    10 per launch
    By
    2 quarters

    Steps

    1. 1

      Template

      Model page.

      Graphic Designer 2 weeks

    2. 2

      Tiers

      Which releases get what.

      Technical Product Marketing Manager 1 week

    3. 3

      Configs

      Tuned defaults.

      Performance team each launch

    4. 4

      Social

      Pre-written posts.

      Social Media Manager each launch

    5. 5

      Offer

      Day-0 assessment.

      Forward Deployed Engineer each launch

    6. 6

      Review

      Results per launch.

      Technical Product Marketing Manager each launch

    First result
    1 quarter
    Cost
    Staff time only
    Owner
    Technical Product Marketing Manager

    Smallest test

    Run the kit for the next major release.

    Why it works

    Prepare a tiered kit: a model page template (how to serve it on vLLM and on Inferact, tuned configs, first benchmark), social posts, partner copy and a day-0 assessment offer. Ship it within hours of each release.

    Being the first credible answer on release day captures high-intent teams.

    Risks and how to handle them

    • Release dates slip

      Keep kits ready to fill.

    Evidence

    • Your websiteHomepage: "From the creators of vLLM. Frontier inference you control."; features include day-zero models, autoscaling, "Frontier Tuning", endpoints and metering, observability, "Forward-deployed Inference Engineers" and SOC 2; "vLLM is optimized for everyone. Inferact is optimized for you." The main call to action, "Book a meeting", opens a Typeform; the footer says "Tell us a model to serve. We'll help optimize it. Talk to an engineer".Source
    • Your websiteHomepage proof numbers: "464 tok/s bs=1 decode on Kimi-K3 on launch day", "709 tok/s/user with first Megakernel on TPU", "18.3M tok / 1$ On Semianalysis AgentX", "#1 On Artificial Analysis, May 2026". Only the 709 tok/s figure links to its source post.Source
    • HiringTechnical Product Marketing Manager (posted September 24, 2026): owns "day-0 launch communications, product positioning, the benchmark narrative, customer proof"; within 60 days owns "website product pages" and "partnership announcements with ecosystem partners such as Google, AMD, and NVIDIA"; within 90 days presents a plan covering "launch tiers, the benchmark program, case study pipeline", and builds a launch process covering "intake, messaging, review, distribution, partner amplification, and post-launch measurement".Source
    • HiringSocial Media Manager (posted September 24, 2026): owns "the daily voice, publishing system, and audience growth strategy for vLLM and Inferact"; "Own the social side of the rebrand launch alongside the website"; a voice guide separating "the vLLM community voice from the Inferact company voice"; a founder program; testing "Hacker News, Instagram, and TikTok"; a "partner amplification network"; and a "repeatable social launch system with tiered templates, asset checklists".Source
  • 09

    The GTM Engineer is asked to mine "ecosystem and community data to identify high-value prospects and warm paths". vLLM has over 3,000 contributors, office hours and events; companies already running vLLM in production show up there.

    High impactMedium effort First result in 1 quarter GTM Engineer

    First action

    List the 50 companies most active in public vLLM issues.

    How to measure it

    Community-sourced meetings

    Now
    –
    Target
    20 a quarter
    By
    2 quarters

    Steps

    1. 1

      Define signals

      Public and consented only.

      GTM Engineer 1 week

    2. 2

      Collect

      Issues, events, office hours.

      GTM Engineer 1 month

    3. 3

      Score

      Size and activity.

      GTM Engineer 1 week

    4. 4

      Hand off

      Context per account.

      GTM Engineer ongoing

    5. 5

      Outreach

      Engineer-led.

      Forward Deployed Engineer ongoing

    6. 6

      Measure

      Meetings and pipeline.

      Head of Commercial monthly

    First result
    1 quarter
    Cost
    Staff time only
    Owner
    GTM Engineer

    Smallest test

    Score the top 50 accounts by hand.

    Why it works

    Build an account list from public signals (company affiliations in contributions and issues, event and office-hours registrations with consent, meetup attendance), score by size and activity, and hand the top accounts to sales with the context.

    Teams already running vLLM are the warmest buyers for a vLLM production platform.

    Risks and how to handle them

    • Community feels watched

      Use only public or opted-in data; never cold-pitch contributors on project channels.

    Evidence

    • HiringGTM Engineer (posted September 29, 2026): "build the commercial operating system for Inferact from the ground up", where "CRM structure, account intelligence, standardized workflows" are "still being defined"; early projects include "a clean commercial source of truth" and "mining Inferact's unique ecosystem and community data to identify high-value prospects and warm paths".Source
    • Your websiteNews: Google Cloud and Inferact announced "a joint engineering and go-to-market partnership to make TPU a first-class citizen in vLLM" (September 14, 2026); vLLM "supports more than 5,000 model architectures, counts a community of over 3,000 contributors"; an AMD collaboration on AMD Instinct GPUs (October 2026); full Kimi K3 vendor verification (August 2026). vLLM is "an Apache 2.0 project governed by the PyTorch foundation".Source
    • vllm.aiOn vllm.ai, Inferact appears only as the "Slack Sponsor" in the sponsors list, with no link; the project site runs events, office hours, a releases page and a vLLM Conference 2026 page.Source
  • 10

    Inferact has quotes from Applied Compute, MiniMax, Prime Intellect, the Kimi team and AMD, but no case studies. The TPMM owns "customer proof" and a "case study pipeline".

    High impactMedium effort First result in 1 quarter Technical Product Marketing M…

    First action

    Ask two quoted customers for a 30-minute interview.

    How to measure it

    Case studies

    Now
    –
    Target
    3
    By
    2 quarters

    Steps

    1. 1

      Ask

      Two customers.

      Head of Commercial 1 week

    2. 2

      Interview

      Numbers and story.

      Technical Product Marketing Manager 2 weeks

    3. 3

      Write

      Short format.

      Technical Product Marketing Manager 2 weeks

    4. 4

      Approve

      Customer sign-off.

      Head of Commercial 2 weeks

    5. 5

      Design

      Page and PDF.

      Graphic Designer 1 week

    6. 6

      Use

      Outbound and decks.

      Head of Commercial ongoing

    First result
    1 quarter
    Cost
    Staff time only
    Owner
    Technical Product Marketing Manager

    Smallest test

    Publish one case study.

    Why it works

    Write three short case studies with before-and-after numbers (tokens per dollar, latency, rollouts per GPU hour), deployment model and time to production.

    Specific results from known AI teams are the strongest proof for the next buyer.

    Risks and how to handle them

    • Customers decline numbers

      Use ranges or relative gains.

    Evidence

    • Your websiteCustomer and partner quotes on the homepage from Applied Compute, MiniMax, Prime Intellect ("In large-scale RL, most of the compute goes to inference for sampling rollouts. Inferact helps us optimize inference to get more rollouts per GPU hour"), the Kimi team at Moonshot AI and AMD; no published case studies.Source
    • HiringTechnical Product Marketing Manager (posted September 24, 2026): owns "day-0 launch communications, product positioning, the benchmark narrative, customer proof"; within 60 days owns "website product pages" and "partnership announcements with ecosystem partners such as Google, AMD, and NVIDIA"; within 90 days presents a plan covering "launch tiers, the benchmark program, case study pipeline", and builds a launch process covering "intake, messaging, review, distribution, partner amplification, and post-launch measurement".Source
    • Your websiteThe sitemap has 10 URLs: home, careers, news, privacy, terms and five news posts. Inferact Platform ("Run inference on your compute") and Inferact Cloud ("Run inference on our compute") exist only as homepage sections; /platform, /cloud, /pricing and /benchmarks return 404. There is no documentation, pricing or case-study page.Source
  • 11

    Inferact has a joint engineering and go-to-market partnership with Google Cloud and an AMD collaboration, and the Enterprise Partnerships Manager owns these relationships. Announcements alone do not create pipeline.

    High impactMedium effort First result in 2 quarters Enterprise Partnerships Manag…

    First action

    Agree a lead-sharing process with one partner.

    How to measure it

    Partner meetings

    Now
    –
    Target
    15 a quarter
    By
    2 quarters

    Steps

    1. 1

      Agree terms

      Lead sharing.

      Enterprise Partnerships Manager 1 month

    2. 2

      Pages

      One per partner.

      Graphic Designer 2 weeks

    3. 3

      Offer

      Joint assessment.

      Forward Deployed Engineer 1 week

    4. 4

      Webinar

      Quarterly.

      Product Marketing Manager quarterly

    5. 5

      Field kit

      For partner sellers.

      Technical Product Marketing Manager 2 weeks

    6. 6

      Measure

      Meetings by partner.

      GTM Engineer monthly

    First result
    2 quarters
    Cost
    Staff time only
    Owner
    Enterprise Partnerships Manager

    Smallest test

    Launch one partner page.

    Why it works

    For each hardware partner, build a "vLLM on X with Inferact" page with tuned results, a joint offer and a lead form shared with the partner's field team, and run one joint webinar a quarter.

    Partner field teams meet buyers choosing hardware, which is when they choose an inference stack.

    Risks and how to handle them

    • Partner neutrality

      Keep pages factual and parallel across partners.

    Evidence

    • Your websiteNews: Google Cloud and Inferact announced "a joint engineering and go-to-market partnership to make TPU a first-class citizen in vLLM" (September 14, 2026); vLLM "supports more than 5,000 model architectures, counts a community of over 3,000 contributors"; an AMD collaboration on AMD Instinct GPUs (October 2026); full Kimi K3 vendor verification (August 2026). vLLM is "an Apache 2.0 project governed by the PyTorch foundation".Source
    • HiringEnterprise Partnerships Manager (posted September 29, 2026): commercial owner for partnerships with "major GPU and semiconductor partners, including organizations such as NVIDIA, AMD, and Intel", and over time reaching out to "emerging accelerator companies".Source
    • HiringTechnical Product Marketing Manager (posted September 24, 2026): owns "day-0 launch communications, product positioning, the benchmark narrative, customer proof"; within 60 days owns "website product pages" and "partnership announcements with ecosystem partners such as Google, AMD, and NVIDIA"; within 90 days presents a plan covering "launch tiers, the benchmark program, case study pipeline", and builds a launch process covering "intake, messaging, review, distribution, partner amplification, and post-launch measurement".Source
  • 12

    The PMM role exists to turn "a high volume of conference and ecosystem demand" into a GTM motion, and the designer will build conference sites for vLLM Conference US and China. Badge scans and talk attendees are lost without a shared workflow.

    High impactMedium effort First result in 1 quarter Product Marketing Manager

    First action

    Set up the workflow for the next conference.

    How to measure it

    Meetings per event

    Now
    –
    Target
    10
    By
    2 quarters

    Steps

    1. 1

      Event page

      Template.

      Graphic Designer 1 week

    2. 2

      Capture

      QR and form into CRM.

      GTM Engineer 1 week

    3. 3

      Follow-up

      48-hour SLA.

      Forward Deployed Engineer ongoing

    4. 4

      Talk tracks

      Booth and speakers.

      Technical Product Marketing Manager 1 week

    5. 5

      Report

      Per event.

      GTM Engineer each event

    6. 6

      Decide

      Which events to repeat.

      Product Marketing Manager quarterly

    First result
    1 quarter
    Cost
    Event budgets
    Owner
    Product Marketing Manager

    Smallest test

    Run it at the next event.

    Why it works

    Give every event an Inferact page, a QR to the assessment offer, consent-based lead capture into the CRM, and a 48-hour engineer follow-up. Report pipeline per event.

    Events are already happening; the system makes them measurable.

    Risks and how to handle them

    • Community events are not sales events

      Capture only with consent; keep project events neutral.

    Evidence

    • HiringProduct Marketing Manager (posted July 23, 2026): "turn a high volume of conference and ecosystem demand into a proactive GTM motion"; owns "conferences, partner events, hosted events, landing pages, speaker coordination, booth needs, collateral, follow-up workflows".Source
    • HiringGraphic Designer (posted September 24, 2026): "Own every blog diagram, partnership announcement kit, social card, and preview image that ships"; build Webflow "event pages, partnership announcement pages, blog templates"; later own "product pages, the public benchmark dashboard home, and conference sites for vLLM Conference US and China".Source
    • HiringGTM Engineer (posted September 29, 2026): "build the commercial operating system for Inferact from the ground up", where "CRM structure, account intelligence, standardized workflows" are "still being defined"; early projects include "a clean commercial source of truth" and "mining Inferact's unique ecosystem and community data to identify high-value prospects and warm paths".Source
    • vllm.aiOn vllm.ai, Inferact appears only as the "Slack Sponsor" in the sponsors list, with no link; the project site runs events, office hours, a releases page and a vLLM Conference 2026 page.Source
  • 13

    The Social Media Manager will run a founder program with ghostwriting and an approval loop. Founders who built vLLM are the most credible voices in inference.

    Medium impactMedium effort First result in 1 quarter Social Media Manager

    First action

    Agree the cadence and approval loop.

    How to measure it

    Founder followers

    Now
    –
    Target
    +50%
    By
    2 quarters

    Steps

    1. 1

      Cadence

      Weekly per founder.

      Social Media Manager 1 week

    2. 2

      Topics

      Backlog of 20.

      Social Media Manager 1 week

    3. 3

      Draft

      Ghostwriting.

      Social Media Manager weekly

    4. 4

      Approve

      24-hour loop.

      Co-Founder and CEO weekly

    5. 5

      Long form

      Monthly.

      Developer Relations Engineer monthly

    6. 6

      Measure

      Reach and site visits.

      GTM Engineer monthly

    First result
    1 quarter
    Cost
    Staff time only
    Owner
    Social Media Manager

    Smallest test

    Publish one post per founder this week.

    Why it works

    One technical post per founder per week (lessons from serving, hardware notes, launch context), with a 24-hour approval loop, and a monthly long-form piece suited to Hacker News.

    Engineers follow builders more than brands; founder reach compounds into hiring and pipeline.

    Risks and how to handle them

    • Founder time

      Keep approvals under ten minutes.

    Evidence

    • HiringSocial Media Manager (posted September 24, 2026): owns "the daily voice, publishing system, and audience growth strategy for vLLM and Inferact"; "Own the social side of the rebrand launch alongside the website"; a voice guide separating "the vLLM community voice from the Inferact company voice"; a founder program; testing "Hacker News, Instagram, and TikTok"; a "partner amplification network"; and a "repeatable social launch system with tiered templates, asset checklists".Source
    • News coverageInferact launched in January 2026 with a $150 million seed round at an $800 million valuation, co-led by Andreessen Horowitz and Lightspeed, to commercialize vLLM.Source
    • Your websiteNews: Google Cloud and Inferact announced "a joint engineering and go-to-market partnership to make TPU a first-class citizen in vLLM" (September 14, 2026); vLLM "supports more than 5,000 model architectures, counts a community of over 3,000 contributors"; an AMD collaboration on AMD Instinct GPUs (October 2026); full Kimi K3 vendor verification (August 2026). vLLM is "an Apache 2.0 project governed by the PyTorch foundation".Source
  • 14

    Inferact sells "maximum tokens per dollar" and DevRel will teach latency versus throughput trade-offs. Buyers need to see the economics for their own model and traffic.

    High impactMedium effort First result in 2 quarters Developer Relations Engineer

    First action

    Sketch the inputs and outputs.

    How to measure it

    Calculator-to-assessment rate

    Now
    –
    Target
    5%
    By
    2 quarters

    Steps

    1. 1

      Design inputs

      Model, hardware, traffic.

      Developer Relations Engineer 2 weeks

    2. 2

      Model costs

      From benchmark data.

      Performance team 1 month

    3. 3

      Build

      Web tool.

      Developer Relations Engineer 1 month

    4. 4

      Explain

      Linked deep dives.

      Developer Relations Engineer 1 month

    5. 5

      Launch

      Social and HN.

      Social Media Manager 1 week

    6. 6

      Measure

      Uses and assessments.

      GTM Engineer monthly

    First result
    2 quarters
    Cost
    Engineering time
    Owner
    Developer Relations Engineer

    Smallest test

    Ship a spreadsheet version first.

    Why it works

    Build a calculator: model, hardware, traffic shape and latency target in; estimated GPUs, tokens per dollar and the configuration to try out, with a link to the assessment offer.

    A useful tool earns links and repeat visits and qualifies buyers by the numbers they enter.

    Risks and how to handle them

    • Estimates questioned

      Show assumptions and ranges.

    Evidence

    • HiringDeveloper Relations Engineer: "make vLLM the default way developers understand, build, and scale AI inference" through deep dives, demos, tutorials and workshops on "KV cache, continuous batching, prefix caching, prefill and decode, quantization, GPU serving, latency versus throughput". Forward Deployed Engineer (posted October 2, 2026): deploys vLLM-powered inference in customer environments, including "enterprise, regulated, security-sensitive, or bring-your-own-cloud environments", and affects "customer time-to-value".Source
    • Your websiteHomepage: "From the creators of vLLM. Frontier inference you control."; features include day-zero models, autoscaling, "Frontier Tuning", endpoints and metering, observability, "Forward-deployed Inference Engineers" and SOC 2; "vLLM is optimized for everyone. Inferact is optimized for you." The main call to action, "Book a meeting", opens a Typeform; the footer says "Tell us a model to serve. We'll help optimize it. Talk to an engineer".Source
    • Your websiteHomepage proof numbers: "464 tok/s bs=1 decode on Kimi-K3 on launch day", "709 tok/s/user with first Megakernel on TPU", "18.3M tok / 1$ On Semianalysis AgentX", "#1 On Artificial Analysis, May 2026". Only the 709 tok/s figure links to its source post.Source
  • 15

    The homepage mentions SOC 2 and that "your weights, traffic, and logs stay secure", and FDEs deploy into "regulated, security-sensitive, or bring-your-own-cloud environments". Security reviewers need detail before a pilot.

    Medium impactMedium effort First result in 1 quarter Forward Deployed Engineer

    First action

    List the questions from the last five security reviews.

    How to measure it

    Time to pilot

    Now
    –
    Target
    -25%
    By
    2 quarters

    Steps

    1. 1

      Collect questions

      Recent reviews.

      Forward Deployed Engineer 1 week

    2. 2

      Diagram

      Architecture.

      Graphic Designer 1 week

    3. 3

      Write

      Page and FAQ.

      Forward Deployed Engineer 2 weeks

    4. 4

      Review

      Security lead.

      Security lead 1 week

    5. 5

      Publish

      Link from products.

      Graphic Designer 1 day

    6. 6

      Measure

      Pilot time.

      GTM Engineer quarterly

    First result
    1 quarter
    Cost
    Staff time only
    Owner
    Forward Deployed Engineer

    Smallest test

    Publish the architecture diagram.

    Why it works

    Publish a page on the deployment architecture (what runs in the customer's cluster, what talks to the control plane), data handling, SOC 2 report access and network requirements.

    Answering security questions up front shortens pilots with larger enterprises.

    Risks and how to handle them

    • Oversharing details

      Keep architecture high-level; share more under NDA.

    Evidence

    • Your websiteHomepage: "From the creators of vLLM. Frontier inference you control."; features include day-zero models, autoscaling, "Frontier Tuning", endpoints and metering, observability, "Forward-deployed Inference Engineers" and SOC 2; "vLLM is optimized for everyone. Inferact is optimized for you." The main call to action, "Book a meeting", opens a Typeform; the footer says "Tell us a model to serve. We'll help optimize it. Talk to an engineer".Source
    • HiringDeveloper Relations Engineer: "make vLLM the default way developers understand, build, and scale AI inference" through deep dives, demos, tutorials and workshops on "KV cache, continuous batching, prefix caching, prefill and decode, quantization, GPU serving, latency versus throughput". Forward Deployed Engineer (posted October 2, 2026): deploys vLLM-powered inference in customer environments, including "enterprise, regulated, security-sensitive, or bring-your-own-cloud environments", and affects "customer time-to-value".Source
  • 16

    Prime Intellect says "In large-scale RL, most of the compute goes to inference for sampling rollouts" and that Inferact gets them more rollouts per GPU hour. Post-training teams are a fast-growing segment with a clear metric.

    High impactMedium effort First result in 2 quarters Head of Commercial

    First action

    Ask the RL customer for a rollout-throughput number to publish.

    How to measure it

    RL meetings

    Now
    –
    Target
    10 a quarter
    By
    2 quarters

    Steps

    1. 1

      Get a number

      Customer result.

      Head of Commercial 2 weeks

    2. 2

      Page

      RL offer.

      Technical Product Marketing Manager 2 weeks

    3. 3

      Benchmark

      Rollout sampling.

      Performance team 1 month

    4. 4

      List

      Post-training teams.

      GTM Engineer 2 weeks

    5. 5

      Outreach

      Engineer-led.

      Forward Deployed Engineer ongoing

    6. 6

      Measure

      Pipeline.

      GTM Engineer monthly

    First result
    2 quarters
    Cost
    Staff time only
    Owner
    Head of Commercial

    Smallest test

    Publish the RL page with one quote.

    Why it works

    Build an RL page and offer around rollout throughput, publish one benchmark for rollout sampling, and run outreach to post-training and agent-training teams.

    A segment with one measurable pain point is easier to win than "everyone who serves models".

    Risks and how to handle them

    • Small segment

      Pair it with agentic inference messaging.

    Evidence

    • Your websiteCustomer and partner quotes on the homepage from Applied Compute, MiniMax, Prime Intellect ("In large-scale RL, most of the compute goes to inference for sampling rollouts. Inferact helps us optimize inference to get more rollouts per GPU hour"), the Kimi team at Moonshot AI and AMD; no published case studies.Source
    • Your websiteHomepage proof numbers: "464 tok/s bs=1 decode on Kimi-K3 on launch day", "709 tok/s/user with first Megakernel on TPU", "18.3M tok / 1$ On Semianalysis AgentX", "#1 On Artificial Analysis, May 2026". Only the 709 tok/s figure links to its source post.Source
    • vllm.aiOn vllm.ai, Inferact appears only as the "Slack Sponsor" in the sponsors list, with no link; the project site runs events, office hours, a releases page and a vLLM Conference 2026 page.Source

Big bets 3

Larger plays that can change the trajectory.

  • 17

    Inferact already reports tokens per dollar using public cost assumptions and runs on NVIDIA, AMD and TPU. No neutral, regular source tracks what open-model inference costs across hardware.

    High impactMedium effort First result in 3 quarters Technical Product Marketing M…

    First action

    Define the method with two hardware partners.

    How to measure it

    Citations and links

    Now
    –
    Target
    50
    By
    1 year

    Steps

    1. 1

      Method

      Public and reproducible.

      Technical Product Marketing Manager 1 month

    2. 2

      Data

      Monthly runs.

      Performance team monthly

    3. 3

      Publish

      Index page.

      Graphic Designer 1 month

    4. 4

      Commentary

      Founder note.

      Co-Founder and CEO monthly

    5. 5

      Distribute

      Social and press.

      Social Media Manager monthly

    6. 6

      Measure

      Citations.

      GTM Engineer quarterly

    First result
    3 quarters
    Cost
    Compute and staff time
    Owner
    Technical Product Marketing Manager

    Smallest test

    Publish one month.

    Why it works

    Publish a monthly index of tokens per dollar for popular open models across hardware, with methods and configs, and a short commentary.

    A recurring index becomes the reference journalists, analysts and buyers cite.

    Risks and how to handle them

    • Partner sensitivities

      Agree method publicly and stick to it.

    Evidence

    • Your websiteThe newest post, "Serving agentic workloads with TPU megakernels in vLLM" (October 2026, featured on the news page and the homepage), has an empty og:image and twitter:image, so links shared on X, LinkedIn or Slack show no preview image, and it is missing from the sitemap. Every earlier post has its own share image. Checked twice on October 22, 2026.Source
    • Your websiteNews: Google Cloud and Inferact announced "a joint engineering and go-to-market partnership to make TPU a first-class citizen in vLLM" (September 14, 2026); vLLM "supports more than 5,000 model architectures, counts a community of over 3,000 contributors"; an AMD collaboration on AMD Instinct GPUs (October 2026); full Kimi K3 vendor verification (August 2026). vLLM is "an Apache 2.0 project governed by the PyTorch foundation".Source
    • Your websiteHomepage proof numbers: "464 tok/s bs=1 decode on Kimi-K3 on launch day", "709 tok/s/user with first Megakernel on TPU", "18.3M tok / 1$ On Semianalysis AgentX", "#1 On Artificial Analysis, May 2026". Only the 709 tok/s figure links to its source post.Source
  • 18

    The Partnerships Manager will reach out to "emerging accelerator companies". New chips need vLLM support to be usable, and Inferact's engineers are best placed to deliver it.

    High impactHigh effort First result in 4 quarters Enterprise Partnerships Manag…

    First action

    List the ten accelerator companies closest to production.

    How to measure it

    Signed partners

    Now
    –
    Target
    3
    By
    4 quarters

    Steps

    1. 1

      List

      Ten companies.

      Enterprise Partnerships Manager 2 weeks

    2. 2

      Package

      Scope and pricing.

      Enterprise Partnerships Manager 1 month

    3. 3

      Pilot

      One partner.

      Performance team 2 quarters

    4. 4

      Report

      Compatibility page.

      Technical Product Marketing Manager 1 month

    5. 5

      Announce

      Joint launch.

      Social Media Manager 1 week

    6. 6

      Scale

      Next partners.

      Enterprise Partnerships Manager quarterly

    First result
    4 quarters
    Cost
    Engineering time
    Owner
    Enterprise Partnerships Manager

    Smallest test

    Sign one pilot.

    Why it works

    Offer a paid enablement program: vLLM backend support, tuning, day-0 model coverage and a published compatibility report, with all core work upstreamed.

    It creates revenue, grows vLLM's hardware reach and makes Inferact the gateway for new hardware.

    Risks and how to handle them

    • Community governance

      Upstream through the normal project process.

    Evidence

    • HiringEnterprise Partnerships Manager (posted September 29, 2026): commercial owner for partnerships with "major GPU and semiconductor partners, including organizations such as NVIDIA, AMD, and Intel", and over time reaching out to "emerging accelerator companies".Source
    • Your websiteNews: Google Cloud and Inferact announced "a joint engineering and go-to-market partnership to make TPU a first-class citizen in vLLM" (September 14, 2026); vLLM "supports more than 5,000 model architectures, counts a community of over 3,000 contributors"; an AMD collaboration on AMD Instinct GPUs (October 2026); full Kimi K3 vendor verification (August 2026). vLLM is "an Apache 2.0 project governed by the PyTorch foundation".Source
    • GitHubThe Inferact GitHub organization has 5 public repositories; the top one, tpu-megakernels ("A collection of megakernels for TPU"), has 167 stars and was updated October 15, 2026.Source
  • 19

    Inferact Cloud runs on GPUs Inferact provisions, and day-0 support makes new models available at launch. Developers who want to try a new model on release day have no self-serve way to do it, while Baseten, Fireworks and Together AI sell platform access to developers.

    High impactHigh effort First result in 4 quarters Co Founder and CEO

    First action

    Model the cost of one day-0 endpoint.

    How to measure it

    Endpoint-to-contract rate

    Now
    –
    Target
    3%
    By
    4 quarters

    Steps

    1. 1

      Cost model

      Per endpoint.

      Finance 1 month

    2. 2

      Metering

      Built-in metering.

      Engineering 1 quarter

    3. 3

      Billing

      Card sign-up.

      Engineering 1 month

    4. 4

      Launch

      With next day-0 model.

      Technical Product Marketing Manager 1 week

    5. 5

      Route

      Heavy users to sales.

      GTM Engineer ongoing

    6. 6

      Review

      Margin and conversion.

      Co-Founder and CEO quarterly

    First result
    4 quarters
    Cost
    Compute costs
    Owner
    Co Founder and CEO

    Smallest test

    Offer one endpoint to 50 developers.

    Why it works

    Offer metered endpoints for a short list of day-0 models with a credit card, and route heavy users to the platform team.

    Launch-day developers become a funnel into enterprise deals, and usage data shows which teams are growing.

    Risks and how to handle them

    • Competing with partners

      Keep endpoints limited to day-0 models.

    Evidence

    • Your websiteHomepage: "From the creators of vLLM. Frontier inference you control."; features include day-zero models, autoscaling, "Frontier Tuning", endpoints and metering, observability, "Forward-deployed Inference Engineers" and SOC 2; "vLLM is optimized for everyone. Inferact is optimized for you." The main call to action, "Book a meeting", opens a Typeform; the footer says "Tell us a model to serve. We'll help optimize it. Talk to an engineer".Source
    • Competitor researchRivals' own homepages: Baseten, "the fastest, most reliable inference platform"; Fireworks, "training and inference platform" for "your specialized intelligence"; Together AI, "The AI Native Cloud" for "inference, fine-tuning, and GPU clusters". RadixArk, the company behind SGLang, was reported to raise at a $400 million valuation the day before Inferact launched.Source
    • Your websiteThe sitemap has 10 URLs: home, careers, news, privacy, terms and five news posts. Inferact Platform ("Run inference on your compute") and Inferact Cloud ("Run inference on our compute") exist only as homepage sections; /platform, /cloud, /pricing and /benchmarks return 404. There is no documentation, pricing or case-study page.Source

Wildcards 2

Unexpected experiments, each with a cheap way to test it.

  • 20

    On vllm.ai, Inferact appears only as the unlinked "Slack Sponsor". Users who want help running vLLM in production have no neutral place to find providers.

    Medium impactLow effort First result in 1 quarter Co Founder and CEO

    First action

    Draft the proposal for the project maintainers.

    How to measure it

    Visits from vllm.ai

    Now
    –
    Target
    500 a month
    By
    2 quarters

    Steps

    1. 1

      Draft

      Neutral criteria.

      Co-Founder and CEO 1 week

    2. 2

      Propose

      Project governance.

      Co-Founder and CEO 1 month

    3. 3

      Listing

      If accepted.

      Social Media Manager 1 week

    4. 4

      Landing

      Inferact page for vLLM users.

      Graphic Designer 1 week

    5. 5

      Measure

      Referral visits.

      GTM Engineer monthly

    6. 6

      Review

      Yearly.

      Co-Founder and CEO yearly

    First result
    1 quarter
    Cost
    $0
    Owner
    Co Founder and CEO

    Smallest test

    Raise it at the next community meeting.

    Why it works

    Propose through the project's governance a page listing companies that offer commercial support or hosting, open to any vendor meeting simple criteria. Inferact would be listed alongside others.

    It helps users and gives Inferact a legitimate path from the project to the company without bending community rules.

    Risks and how to handle them

    • Seen as self-serving

      Make the listing open to all vendors.

    Evidence

    • vllm.aiOn vllm.ai, Inferact appears only as the "Slack Sponsor" in the sponsors list, with no link; the project site runs events, office hours, a releases page and a vLLM Conference 2026 page.Source
    • Your websiteNews: Google Cloud and Inferact announced "a joint engineering and go-to-market partnership to make TPU a first-class citizen in vLLM" (September 14, 2026); vLLM "supports more than 5,000 model architectures, counts a community of over 3,000 contributors"; an AMD collaboration on AMD Instinct GPUs (October 2026); full Kimi K3 vendor verification (August 2026). vLLM is "an Apache 2.0 project governed by the PyTorch foundation".Source
  • 21

    The Social Media Manager and Graphic Designer posts label their milestones "Outside Counsel Alignment & Contracting Intake", "Frontier AI & Regulatory Strategy" and "Scalable Legal Infrastructure". Candidates for brand and content roles notice copy errors.

    Low impactLow effort First result in 1 day Recruiting lead

    First action

    Edit the two posts in Ashby.

    How to measure it

    Reviewed posts

    Now
    –
    Target
    100%
    By
    2 weeks

    Steps

    1. 1

      Edit

      Two posts.

      Recruiting lead 1 hour

    2. 2

      Check others

      All open posts.

      Recruiting lead 1 hour

    3. 3

      Template

      Clean milestone format.

      Recruiting lead 1 hour

    4. 4

      Review step

      Second reader.

      Recruiting lead ongoing

    5. 5

      Share

      Updated posts on social.

      Social Media Manager 1 day

    6. 6

      Repeat

      Quarterly check.

      Recruiting lead quarterly

    First result
    1 day
    Cost
    $0
    Owner
    Recruiting lead

    Smallest test

    Edit both posts today.

    Why it works

    Replace the headings with ones that match each role, and add a quick review step before posts go live.

    Job posts are the first piece of marketing a candidate sees, and these roles are about care for detail.

    Risks and how to handle them

    • None material

      Small change.

    Evidence

    • HiringThe Social Media Manager and Graphic Designer posts label their 30/60/90-day milestones "Outside Counsel Alignment & Contracting Intake", "Frontier AI & Regulatory Strategy" and "Scalable Legal Infrastructure", headings that do not match the work listed under them.Source
    • HiringSocial Media Manager (posted September 24, 2026): owns "the daily voice, publishing system, and audience growth strategy for vLLM and Inferact"; "Own the social side of the rebrand launch alongside the website"; a voice guide separating "the vLLM community voice from the Inferact company voice"; a founder program; testing "Hacker News, Instagram, and TikTok"; a "partner amplification network"; and a "repeatable social launch system with tiered templates, asset checklists".Source
    • HiringGraphic Designer (posted September 24, 2026): "Own every blog diagram, partnership announcement kit, social card, and preview image that ships"; build Webflow "event pages, partnership announcement pages, blog templates"; later own "product pages, the public benchmark dashboard home, and conference sites for vLLM Conference US and China".Source

03

How Programmable can help

Where an idea in this playbook is something Programmable builds or runs, it's listed here with what we'd do. Everything else is yours to run.

ProgrammableTalk

Fill the events in this plan

These ideas are events. Talk gets registrants for them and tells you who's coming and what their company does before the day.

Plan an event (opens programmable.talk in a new tab)
More example playbooks and reviews
ProgrammableDigital

Show Inferact’s product working

Turn screenshots or the website into an animated product demo in its own brand, for the homepage, launch posts and outreach in this playbook.

Make a demo (opens programmable.digital in a new tab)

More growth playbooks

Other companies we have researched, each read from what they are hiring for.

  • VKAPS IT Solutions growth strategyAn Indore-based software and AI agency (founded 2010 as Future Technologies) that builds MVPs, SaaS products, web and mobile apps and AI…22 growth ideas
  • Nautilus growth strategyAn AI-native platform for car washes that runs on the wash's existing POS: online membership sales and checkout, a member portal, CRM and…22 growth ideas
  • KHAITE growth strategyA New York luxury house founded in 2016 by creative director Catherine Holstein, making womenswear, leather goods, shoes and accessories.21 growth ideas
  • Nintex growth strategyA process automation and orchestration platform: process management (now Nintex Processes), workflow, apps, forms, document generation…21 growth ideas
  • Juicebox growth strategyAn AI recruiting platform: natural-language search across 800M+ profiles from 30+ sources, a recruiting CRM and email outreach, and…21 growth ideas
  • Hawthorne Residential Partners growth strategyA Greensboro-based multifamily operator and third-party property manager: 64,000+ units across 275+ properties in eight states, with…20 growth ideas
  • Forgepoint growth strategyFrank, an AI inside sales rep for manufacturers that reads RFQs and drawings, chases missing specs, drafts quotes from the shop's own cost…20 growth ideas
  • onX growth strategyGPS mapping apps for outdoor recreation, onX Hunt, Offroad, Backcountry and Fish (plus TroutRoutes), with land ownership maps, offline…22 growth ideas

See every playbook and website review

Independent research. Programmable wrote this playbook from public sources, as they stood on October 23, 2026, and cites them throughout. Inferact did not commission, review or endorse it, and Programmable is not affiliated with Inferact. Company and product names are trademarks of their owners, used only to identify them. The ideas are suggestions and opinion, not professional, legal or financial advice, and not statements about how Inferact runs its business.

Something wrong, or do you represent Inferact? Email [email protected] and we will correct it or take it down.

30/60/90 roadmap

What to focus on in each month, and the ideas that carry it.

  1. Days 1 to 307 ideas

    Launch-post basics, sourced homepage numbers, product pages, a qualifying form, the assessment offer and the open-source FAQ; fix the job posts.

  2. Days 31 to 6011 ideas

    Benchmark dashboard, day-0 kit, community-sourced accounts, case studies, partner pages, event follow-up, founder program, calculator, security page and the RL segment; propose the vllm.ai support page.

  3. Days 61 to 903 ideas

    Commit to the cost index, the accelerator program and day-0 endpoints.

The research behind the plan

3 research areas, each summarised in a few lines. Open any of them for the full detail.

Customers

What customers say

Customers value support from the engineers who build vLLM. The most cited problem: GPU cost.

  • Day-0 support for new models
  • Latency versus throughput trade-offs
  • In their words: “tokens per dollar”

Market

Competitive landscape

4 competitors compared: Baseten, Fireworks AI, Together AI and RadixArk (SGLang).

  • Gap to exploit: Reproducible public inference benchmarks
  • Baseten: Not the vLLM maintainers
  • Fireworks AI: Not the vLLM maintainers

Hiring signals

Inferact is building its first marketing and GTM team.

  • Technical Product Marketing Manager: Launches, product pages, benchmark program. Ideas 1, 2, 3, 7, 8, 10.
  • Social Media Manager: Launch system and founder program. Ideas 1, 6, 13.
  • Graphic Designer: Every preview image; benchmark dashboard. Ideas 1, 7.

From the team that wrote this playbook

All products
  • ProgrammableAgencyWorking AI systems in days, not quarters (opens programmable.agency in a new tab)
  • ProgrammableInfoAn agent that finds, researches and asks firstNew accounts get 100,000 free enrichment credits with code BETA100K (opens programmable.info in a new tab)
  • ProgrammableDigitalScreenshots in, an animated product demo out (opens programmable.digital in a new tab)
  • ProgrammableWatchKnow the day a supplier falls out of compliance (opens programmable.watch in a new tab)
  • ProgrammableTalkGet registrants for your events, and know who’s coming (opens programmable.talk in a new tab)
  • ProgrammableMediaAn AI note taker, interpreter and webinar host for your calls (opens programmable.media in a new tab)
  • ProgrammablePageText what you need, your customer gets a page that works (opens programmable.page in a new tab)
  • ProgrammableCommunityForums with labelled AI members, never empty (opens programmable.community in a new tab)
ProgrammableMarketing
ExamplesLeaderboardsProductsTermsPrivacyCookiesDPALegalContact

© 2026 AI-Powered Growth Intelligence