- 01Relaunch until it lands
- 02Turn your best customer into a sales document
- 03Seed micro-creators with a code, not a media buy
- 04Sell the gatekeeper, not the individual
01 Snapshot
The one-glance read on who they are and how they grow. Each point is verifiable from the receipts above.
An AI research agent that reads hundreds of scientific papers to answer complex literature-search queries, aimed at the researchers who rely on PubMed and Google Scholar, built by two MIT physics PhDs, Joshua Ramette and Tom Hartke, out of Y Combinator's Summer 2024 batch.
Launch-led, with the initial user base built through repeated Hacker News launches that later converted into enterprise and institutional deals.
≈240,000 monthly visits (down 4% over the last 3 months) as of 2026-08-01, a $500K pre-seed closed 2024-09-25, and a GSK deployment reaching 1,000+ scientists daily, per a 2026-05 trade-press feature on GSK's AI stack.
Undermind didn't find its audience with one clever launch, it took three failed Show HN attempts before the fourth, official YC launch broke through, and the same "keep resubmitting until it lands" persistence now shows up as an enterprise motion: one large, well-documented customer (GSK) doing the selling instead of a content or ad engine.
02 How it unfolded
The order the channels came online. Sequence is strategy: what they did first, and what they layered on once demand existed.
03 Traffic & channel mix
Estimated demand, the channel split behind it, and the keywords and referrers doing the work. Directional modeling, not audited analytics.
Undermind draws an estimated 240,000 monthly visits, down 4% over the last 3 months as of the 2026-08-01 observation date. No per-channel breakdown, keyword list, or referrer data was available in this pull, so the channel mix behind that total cannot be stated here without guessing; the launch, institutional, and enterprise evidence below is the more reliable read on how that traffic actually arrives.
04 Channel map
Not traffic share. How much weight the growth system actually puts on each channel, with a one-line read on the role it plays.
What drove growth
05 The motions doing the work
For founder-led SaaS the breakdown shifts from ads to traction: where the first users came from, how the founder grows it in the open, and the compounding organic surface.
Undermind's traction traces to a three-attempt Hacker News campaign that finally broke through as an official YC launch, then converted into enterprise reach through a single flagship customer relationship.
Co-founder Tom Hartke posted three separate Show HN submissions himself: "Deep search of all ML papers" (2024-01-24, 109 points, 25 comments), "Deep search engine for science" (2024-03-26, 3 points, 1 comment), and "Undermind, Deep scientific paper search with adaptive LLMs" (2024-04-02, 49 points, 14 comments), each a different framing of the same product.
Co-founder Joshua Ramette's "Launch HN: Undermind (YC S24)" on 2024-07-25 hit 292 points and 126 comments, by far the strongest single result and timed to their official YC batch launch rather than a self-initiated Show HN.
Dig deeper+2▾
A $500K pre-seed led by Y Combinator, with Unpopular Ventures, Pioneer Fund, SBXi, and FundersClub participating, closed 2024-09-25, roughly two months after the Launch HN moment.
Growth shifted from a consumer-style launch to institutional selling: Singapore Management University opened an institutional trial by 2024-08-21, and GSK later deployed Undermind as the research-evidence layer of its internal "Cogito Forge" AI-scientist system, reaching over 1,000 scientists daily.
Undermind's build-in-public motion runs through Hacker News under the founders' own names rather than through a personal social following.
Tom Hartke and Joshua Ramette each posted launches under their own handles (tomhartke, jramette) instead of a brand account, making the repeated Show HN attempts covered above their main visible founder-voice channel.
The homepage links to Ramette's X account and the company's LinkedIn page, but no post activity, cadence, or follower counts were captured in the evidence, so the reach of either channel cannot be sized here.
Dig deeper+1▾
There is no evidence of MRR-milestone threads, founder Reddit posts, or a build-in-public content cadence. Growth instead reads as institution- and customer-led: the SMU library relationship and the GSK deployment stand in for the personal-audience engine a solo-founder SaaS company would typically lean on.
Undermind's content investment is concentrated in one flagship asset rather than a compounding search-content system.
The GSK deployment is published as a co-authored case study with GSK's SVP and Global Head of AI/ML, Kim Branson, functioning as a credibility asset for enterprise buyers more than a search-traffic driver.
No Meta or Google ad library evidence surfaced for the domain. The Instagram and TikTok posts referenced above instead run on personal referral links and discount codes (one creator promotes a "30% off" code), a lightweight affiliate-style lever recruited through individual creators rather than a paid-media budget.
Dig deeper+1▾
There is no blog, comparison-page network, or documentation hub in the evidence, which is consistent with a company whose acquisition still runs through direct launches and institutional relationships rather than programmatic search content, a gap a competitor could exploit before Undermind builds one.
06 The playbook
The channels are not separate. They are one system where each stage feeds the next. Here is the read, then the plays to run tomorrow.
The inferred playbook
Undermind's loop starts with a technical launch that earns an initial research-user base, then compounds through institutional and enterprise relationships instead of paid acquisition or content.
The Hacker News launches and small-scale creator referral posts bring in individual researchers at near-zero cost, evidenced by the 292-point Launch HN and the sub-2,700-engagement creator posts covered above.
A freemium funnel (free tier, then $16 to $20/month Pro billed as reported by third-party pricing pages and the company's own pricing page, then $15/seat/month Team with a 5-seat minimum) lets individual researchers self-upgrade, while SMU and GSK show the same product sold top-down as an institutional or enterprise license.
Dig deeper+3▾
The GSK case study is a reusable enterprise-sales asset, backed by an on-record SVP quote, that can be pointed at other pharma and R&D buyers without new content production; with roughly $440K in reported revenue on a ~4-5 person team as of September 2025, the company appears to convert a small number of large accounts rather than high-volume self-serve traffic.
There is no always-on paid channel, no comparison or SEO content hub, and no personal-founder audience to fall back on if the launch-and-case-study pipeline slows.
the free-to-paid conversion rate off the freemium tier, per-seat expansion revenue inside Team and Enterprise accounts once SMU-style institutional deals are signed, and any traffic or revenue figure more recent than the September 2025 revenue report and the 2026-08-01 traffic pull.
Relaunch until it lands
The proofUndermind's founders posted three separate Show HN attempts, at 109, 3, and 49 points, before the fourth submission, an official Launch HN tied to their YC batch, hit 292 points and 126 comments.
The adaptationPick one community-launch venue in your category (Show HN, a relevant subreddit, a niche Discord or Slack) and ship two or three distinctly different framings of the same product over successive weeks, tracking which specific hook and title actually pulls comments, not just views, then double down on that framing for any bigger launch moment later.
Cost: $0 · Time to signal: days · Works pre-PMF: yes
Turn your best customer into a sales document
The proofUndermind's GSK deployment became a published case study co-authored with GSK's own SVP of AI/ML, the enterprise-credibility asset referenced above.
The adaptationOnce you land one customer who is meaningfully large or recognizable in your category, ask their champion to co-sign a short, quote-driven case study naming their title and specific outcome, then lead with that document, not a generic pitch deck, in every subsequent outbound email to similar accounts.
Cost: $0 · Time to signal: weeks to months · Works pre-PMF: conditional, only once you have one real, named customer willing to go on record.
Seed micro-creators with a code, not a media buy
The proofUndermind's Instagram and TikTok posts run on personal referral links and a "30% off" discount code rather than paid brand ad spend.
The adaptationRecruit five to ten small creators in your niche with a personal discount code and a revenue share instead of an upfront fee, and track signups by code so you know within weeks which specific creator's audience actually converts before you spend real money on any of them.
Cost: under $500 · Time to signal: weeks · Works pre-PMF: yes
Sell the gatekeeper, not the individual
The proofUndermind's SMU Libraries deal turned one institutional buyer into 100+ active monthly users, the seat-license channel covered above, running alongside its GSK enterprise deployment.
The adaptationIdentify the single buyer in your category who can flip on access for a whole group at once, a library, a corporate L&D budget, a community organizer, and pitch that one person a pilot or trial instead of chasing individual signups one at a time; the pitch should lead with how many end users the pilot unlocks, not the per-seat price.
Cost: $0 to under $500 · Time to signal: months · Works pre-PMF: conditional, only if the product is genuinely useful to a group, not just a single user.
Cap usage, not features, on the free tier
The proofUndermind's free tier limits deep searches while its Pro tier unlocks roughly 10x the usage at $16-20/month, a usage-based upgrade rather than a feature-gated one.
The adaptationPut an explicit, visible cap on your product's single highest-cost core action (searches, generations, exports) and price the paid tier as "more of the same action" rather than bundling unrelated features into it, so the upgrade decision is a one-line math problem for the user, not a feature comparison.
Cost: $0 · Time to signal: weeks · Works pre-PMF: conditional, only once the core action is proven valuable enough that hitting the cap is a real frustration, not an annoyance that drives users away. Not transferable at an earlier stage: the GSK-style enterprise case study and the SMU-style institutional license both depend on already having landed one credible, named customer of real scale, something a genuinely pre-product-market-fit founder usually has to earn through plays 1 and 3 first before play 2 or 4 becomes available to them.
You just read how Undermind does it. Tomorrow it is someone else.
Every morning we take one company that is actually growing and break down where its customers come from: the ads still running after a year, the channel doing the real work, and the play you can run this week.
Systemaic · directional intelligence. Traffic, spend, and reach figures are SimilarWeb-style estimates and qualitative reads of public data, not audited numbers. Built on real public receipts.
// teardowns with a similar GTM motion
sources & receipts (27)
Launch archives: Hacker News: Launch HN: Undermind (YC S24) – AI agent for discovering sci · Hacker News: Show HN: Deep search of all ML papers · Hacker News: Show HN: Undermind – Deep scientific paper search with adapt · Hacker News: The Undermind Episode 33 – A Podcast about Starcraft AI Deve · Hacker News: The Undermind Episode 31 – A Podcast about Brood War AI · Hacker News: Undermind Episode 21 – A Podcast about Starcraft AI
Researched facts: Founders and founding year: Undermind was founded in 2023 by Joshua Ramette (CEO) and Tom · Founder Joshua Ramette background: CEO Joshua Ramette did his PhD at MIT studying quantum · Founder Tom Hartke background: CTO Tom Hartke studied physics at Princeton then earned a q · Y Combinator batch: Undermind went through Y Combinator's Summer 2024 (S24) batch. · Pre-seed funding round: Undermind raised $500,000 in a Pre-Seed round on 2024-09-25. · Pre-seed investors: The $500K pre-seed round was led by Y Combinator, with participation f · Total employee count: Undermind has approximately 5 total employees. · GSK enterprise adoption: GSK deployed Undermind enterprise-wide, with over 1,000 GSK scien · SMU Libraries institutional subscription start date: Singapore Management University (SMU) · SMU active user count (as of ~2026): Over 100 active users log in and search using Undermi · Pricing - Pro tier (as of April 2026): The Pro plan is priced at $20/month for 30 deep sea · Pricing - Enterprise tier: Undermind offers a custom-priced Enterprise plan with increased
Milestone sources: 2024-04-03: Undermind publicly unveils its AI-powered scientific literature search tool in · 2024-08-20: Undermind launches publicly on Product Hunt as "10-50x better search for scien · 2024-08-21: Undermind enters the academic-library channel: Singapore Management University · 2025-09: Undermind reaches roughly $440K in revenue with a 4-person team, per founder-inte · 2026-05: GSK's proprietary "Cogito Forge" AI-scientist platform is revealed to run on Unde
Community threads: Academic journal (Journal of the Canadian Health Libraries Association): The same review f · Katina Magazine (librarian trade publication): Librarian-blogger Aaron Tay (a widely-read · Substack (Aaron Tay's Musings about Librarianship): Tay uses Undermind as his personal yar · Substack (Aaron Tay's Musings about Librarianship): Despite liking the results, Tay critic
Traffic, spend, and revenue figures are estimates as noted in the report; the links above are the primary public artifacts.