Skip to content

asmuelle/claimwatch

Repository files navigation

ClaimWatch

CI

Patent and IP radar for IP-heavy startups: a weekly agent over USPTO/EPO/WIPO bulk feeds maintains living dossiers per technology area and competitor — alerting on new filings in your claim space, claim amendments, assignments, and lapses, with every assertion checkable against the canonical document.

Category: LLM wiki / auto-research (living documents + delta alerts, à la Karpathy)

Concept

Patent and IP radar for IP-heavy startups: a weekly agent over USPTO/EPO/WIPO bulk feeds maintains living dossiers per technology area and competitor — alerting on new filings in your claim space, claim amendments, assignments, and lapses, with every assertion checkable against the canonical document.

Target User

Founders, CTOs, and in-house counsel at deeptech, biotech, hardware, and AI startups whose valuation rests on IP (fundraising diligence and freedom-to-operate anxiety are board-level issues), plus boutique patent law firms managing many small clients who want a $400/hr associate's monitoring work turned into a reviewable brief.

Auto-Research Mechanic (the living document + delta engine)

Built on free, structured, canonical government data: USPTO full-text and assignment feeds, EPO OPS, WIPO PATENTSCOPE, litigation dockets — refreshed on the official Tuesday/Thursday publication cycle (cadence matches the data). Watchlists defined as CPC classes plus natural-language claim-space descriptions and named competitors/inventors. Cheap-model screening of new publications and amendments; frontier pass updates living dossiers: per-family claim-evolution timelines, assignment changes, prosecution status, fee lapses and expirations. Weekly brief reasons over prosecution history: 'Competitor X's continuation narrowed claim 1 after rejection — this opens design space in Y.' Every assertion links to document, claim number, and publication date — the highest-citation-trust vertical possible because every citation is mechanically checkable.

Product Surface

Web SaaS (dossiers, claim timelines, watchlist management) + weekly email brief as primary touchpoint — patent monitoring is inherently weekly-cadence and the brief gets forwarded to boards and outside counsel. Counsel tier adds PDF brief export formatted for client files.

Why Now (2026 timing)

Patent watch explicitly identified as an under-disrupted niche with quote-only incumbents and zero cheap challengers. LLMs only recently became capable of meaningfully reading claim language and prosecution histories — and the corpus is free, structured, and crawl-unrestricted, so the access constraints hardening across the general web in 2026 don't apply at all.

Proposed Monetization

$149/mo Startup (1 technology-area watchlist, 10 competitors/CPC classes), $399/mo Pro (5 watchlists, litigation-docket monitoring, counsel-ready export), $799/mo Firm (multi-client workspaces, white-label briefs). The entire sub-$1K/mo tier is empty under PatSnap/Clarivate/Questel's $20K+ quote-only floor; COGS is a few dollars per watchlist on free data.

Competition & Gap

PatSnap/Clarivate/Questel (enterprise, quote-only database access plus analyst tooling), Google Patents (free search, zero monitoring or synthesis), keyword-matching alert services. None maintain a living claim-evolution dossier or explain what an amendment means for your design space.

Claimed Moat

(1) Claim-language extraction and prosecution-history reasoning is deep vertical NLP on a corpus horizontal chatbots don't index conversationally — a scheduled ChatGPT task cannot diff claim amendments across a patent family. (2) The longitudinal claim-evolution graph is accumulated state: two years of watching how a rival's claims narrowed is retroactively irreproducible. (3) Law-firm tier creates distribution lock-in as firms standardize client deliverables on the brief format. (4) Provable accuracy in the one research domain where every citation is mechanically verifiable builds a trust brand that survives any incumbent feature launch.


Evaluation (multi-agent adversarial review)

Comparables

  • PatSnap — quote-only, ~$15K/yr single-seat to $40K-$80K+/yr enterprise; $100M ARR (2023, +20% YoY), $352M raised, $1B valuation
  • Questel — ~$750M annual revenue (2025), enterprise IP software and services, PE-owned
  • Clarivate — $2.56B total revenue (2024); divested its IP product group in 2024, signaling incumbent retreat from the segment
  • Patlytics — AI patent platform, $40M Series B (Apr 2026), ~$65M total raised, 40% of Am Law 100 as customers; nearest well-funded threat
  • Solve Intelligence — AI patent drafting/prosecution, $40M Series B (Dec 2025), $55M total, eight-figure ARR up 10x YoY; 60% law firms / 40% corporate IP
  • IamIP — patent search and monitoring from $99/mo; proves a low-cost tier exists but is keyword-alert grade, no claim-evolution synthesis
  • XLPat — free patent alerts (keyword matching); price-anchor pressure at the bottom of the market
  • Boutique IP firm subscription services — $2K-$10K/mo per client; one firm reports $2.4M/yr from 30 subscription clients (validates Firm-tier white-label economics)
  • Patent attorney hourly rates — $200-$800/hr; the inferior 'alternative' startups currently pay for monitoring/FTO work

Tech Stack & Unit Economics

Ingestion: USPTO Open Data Portal bulk delta datasets (daily) + Patent File Wrapper API (free), EPO OPS paid tier (~EUR 1-4K/yr at scale; free 4GB/mo fair-use tier insufficient beyond pilot), WIPO PATENTSCOPE via vendor or constrained scraping, CourtListener/RECAP + metered PACER (or Docket Alarm API) for the litigation tier; cron aligned to the Tue (grants) / Thu (pre-grant pubs) cycle, Temporal or pg_cron workers. Storage: Postgres + pgvector (claim embeddings via voyage-3 or text-embedding-3-large), S3 for raw XML/PDF, claim-version tables per family. Critical design rule: claim diffing is deterministic XML text diff — never LLM-generated. Screening: CPC prefilter then Haiku 4.5 / Gemini Flash relevance classification of ~200-800 docs/week per watchlist (~1-3M tokens/wk, $1-3/wk). Synthesis: Sonnet 4.6 dossier updates + weekly brief with cite-or-omit grounding — every sentence must resolve to a stored doc ID + claim number and is mechanically validated post-generation; Opus-class pass only for flagged prosecution-history reasoning over office-action context (50-150K tokens/family, $3-10/wk). App: Next.js + Postgres SaaS, Resend/Postmark email brief, PDF export for counsel tier. Unit economics: Startup tier COGS ~$20-40/user/mo (screening $5-12, frontier synthesis $12-40 across 4 weekly briefs amortized, embeddings/infra $3-5) against $149 = ~70-85% gross margin; Pro litigation-docket tier adds $30-150/mo PACER/docket exposure and must be metered or it eats margin. Team of 2-3 can ship an MVP in roughly one quarter; the hard 20% is claim-space recall evaluation, which needs a hand-labeled ground-truth set per vertical before any retention claim is believable.

About

Patent & IP radar for IP-heavy startups: a weekly agent over USPTO/EPO/WIPO feeds maintains living dossiers per technology area and competitor — alerting on new filings, claim amendments, assignments, and lapses, every assertion checkable against the source.

Topics

Resources

Security policy

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors

Languages