← All Builds DAY 21 OF 30

PatentFlow

AI-powered intellectual-property lifecycle management for patent attorneys and IP firms — prior-art search, application drafting, filing, prosecution and docketing, infringement detection, and litigation support in a single platform, built on Claude via AWS Bedrock · React 18 + TypeScript + Vite + Tailwind CSS + shadcn/ui (Blue / New York theme) single-page application at getpatentflow.com — six matter-centric modules (search, drafting, filing, prosecution, infringement, litigation), Cognito-authenticated, served via S3 + CloudFront on a custom domain with ACM TLS · AWS Lambda (Node.js 20) + API Gateway v2 (HTTP API) backend — handler set covering matters, AI search, AI drafting, AI analysis, docketing, documents, filing, and a USPTO proxy; an async job pattern (job-enqueue + worker handlers) carries long-running AI calls so the UI never blocks on a model response · Claude Sonnet 4.6 via AWS Bedrock for drafting, analysis, and multi-step reasoning; Claude Haiku 4.5 via AWS Bedrock for classification, routing, and extraction — prompt-cached on production calls with explicit max_tokens per call site and per-model cost routing · USPTO integration layer — six tool wrappers (patent search, patent full-text retrieval, prior-art search, office-action retrieval, PTAB search, litigation search) over the USPTO Open Data Portal, PatentsView, and PTAB APIs, exposed to the AI layer for multi-turn tool execution · DynamoDB single-table data model for matters, documents, and AI sessions; S3 for document and patent-file storage; AI drafting output rendered to DOCX to meet the USPTO text-submission requirement · AWS Cognito user pools for authentication and session management · AWS CDK v2 (TypeScript) on the Silverback VelocityStack base class — auth, data, api, and frontend stacks with mandatory cost-attribution tagging and an auto-provisioned $500/mo budget alarm, deployed to a dedicated AWS account (client-patentflow-prod, us-east-1)
LIVE https://getpatentflow.com

Day 21 of 30 — the Regulated Legal arc reopens

PatentFlow is an AI-powered IP lifecycle platform for patent attorneys. It is live today at getpatentflow.com, with six integrated modules covering the full arc of patent practice — prior-art search, application drafting, filing, prosecution, infringement detection, and litigation support — running on Claude via AWS Bedrock. PatentFlow is the Day 21 launch in the 30-day Velocity run, and it reopens the Regulated Legal arc that Day 10 CounselExpress opened: the same enterprise-tier, privilege-preserving posture, carried into a different corner of legal practice.

The problem

A boutique IP firm runs on four to eight disconnected tools. One product for prior-art search. Another for drafting. A docketing system that talks to neither. A separate platform for infringement analysis. Litigation support somewhere else again.

The fragmentation is not an accident of any one firm’s procurement — it is the shape of the market. Patent prosecution tools ignore litigation. Drafting assistants do not track statutory deadlines. Infringement platforms never connect back to the prosecution record. Every vendor in the category is a point solution, and the firm is left to be the integration layer.

That integration layer is expensive. A mid-size IP firm spends an estimated $200,000–$500,000 a year on its IP technology stack, and integration overhead — the cost of moving the same matter between systems, reconciling the copies, and catching the drift — runs 20–30% of that budget. The average cost to draft and prosecute a single patent application is around $15,000. And every handoff between two tools is a re-keyed copy: a place where a docket date, a claim number, or a priority date can quietly diverge from what the record actually says. In prosecution, a missed deadline is not an inconvenience — it is a malpractice event.

The competitive narrative underneath PatentFlow is one line: five tools is not a strategy. That reframes every buying conversation from a feature comparison to a business-model comparison — and that is the conversation PatentFlow is built to win.

What PatentFlow does — six modules, one workflow

PatentFlow is a single platform with six matter-centric modules. Each module delivers standalone value, and each hands the next a structured record instead of a re-keyed copy.

Module 1 — Prior Art Search. Semantic search across USPTO databases in natural language. The attorney describes the invention; PatentFlow surfaces relevant prior art, with results persisted to the matter and a searchable history.

Module 2 — Application Drafting. Claims and specifications generated from an invention disclosure, with a quality-check engine that flags antecedent-basis problems, claim-numbering errors, and figure-coordination gaps before the attorney reviews the draft. Output renders to DOCX to meet the USPTO text-submission requirement.

Module 3 — Filing Management. Workflows for provisional, non-provisional, PCT, and design applications with automated checklists, so document assembly and form completion happen against a structured matter rather than a folder of loose files.

Module 4 — Prosecution Support. Office actions analyzed by the AI layer with suggested responses; statutory deadlines tracked and docketed so none get missed. This is the module a chief of practice cares about most, because this is where the malpractice exposure lives.

Module 5 — Infringement Detection. Claim charts generated and mapped against potentially infringing products systematically — and because the prosecution history is already in the system, the infringement analysis runs against the real record, not a re-keyed summary of it.

Module 6 — Litigation Support. Claim-construction analysis, damages estimation, cease-and-desist letter drafting, and litigation cost budgeting.

The integration is the product. A firm does not buy six tools from PatentFlow — it stops being the integration layer.

The architecture — on the right side of the privilege line

PatentFlow makes three architectural choices that a patent attorney evaluating it will care about more than any feature.

Inference runs on Claude via AWS Bedrock, not a consumer AI endpoint. This is the enterprise-tier posture Day 10 CounselExpress runs on — and for patent work the stakes are doubled. Patent prosecution is privileged attorney work product. A pre-grant patent application is also a trade secret, and its value depends on not being publicly disclosed before filing — the novelty bar is unforgiving, and a disclosure event before the priority date can compromise patentability. An invention disclosure pasted into a consumer chatbot is arguably exactly that kind of disclosure event. AWS Bedrock keeps the firm’s matters inside an enterprise contractual envelope where inputs are excluded from model training. Same model. Different legal posture. For patent work it is not optional. The line United States v. Heppner drew for consumer-tier AI use and attorney-client privilege has a sharper edge in patent practice, because the firm is protecting not just privilege but the patentability of the client’s invention.

USPTO-integrated by construction. PatentFlow reads directly from the USPTO Open Data Portal, PatentsView, and PTAB through a six-tool integration layer exposed to the AI for multi-turn tool execution. Search, prosecution, and infringement run against the live USPTO record, not a stale internal mirror.

The attorney keeps the judgement layer. PatentFlow does the reading, structuring, and drafting. Every output — claims, specifications, office-action responses, claim charts — is cite-checkable and edit-ready. The platform is built to be supervised by a registered practitioner, not to act autonomously. The practitioner’s signature and professional responsibility are unchanged; what changes is how much mechanical work sits between the practitioner and a clean draft.

Built alongside the customer

PatentFlow was designed with a practicing patent attorney as domain-expert co-founder, with a go-to-market co-founder on the team — the same operator-validated-founder pattern the rest of the Velocity run is built on. The person who knows where antecedent basis breaks, why a missed docket date is a malpractice event, and what a litigation funder actually wants inside a diligence package is at the table, not in a 5%-advisor seat.

The founding team’s names are being held until each founder has explicitly signed off on being named publicly. When that consent is confirmed, a follow-up post will introduce the team. The pattern matters more than the names for now: the regulated-vertical product is defensible because the domain expert is a principal, and that is the through-line from Day 3 PartFoundry and Day 9 CogleGroup to Day 10 CounselExpress to here.

PatentFlow is the second leg of the Regulated Legal arc. Day 10 CounselExpress was the first — AI litigation support for personal-injury attorneys, built on the enterprise-tier posture United States v. Heppner (S.D.N.Y., February 2026) left open when it held that consumer-tier AI use waives attorney-client privilege. PatentFlow carries the same posture into patent practice.

The arc is one argument made twice in two verticals: AI in a regulated profession has to run inside the contractual and architectural envelope the profession’s rules actually require, and a platform that ignores that envelope is not a shortcut — it is a liability the firm inherits. CounselExpress made the argument for litigation files. PatentFlow makes it for invention disclosures and prosecution histories, where the envelope protects not only privilege but the patentability of the invention itself.

Pricing & business model

  • Per-seat SaaS, tiered by firm size and module depth — structured so a solo practitioner and a mid-size boutique can both be on the right plan. The tiers settle after the founding cohort’s feedback is in; PatentFlow is not publishing a self-serve price table during the design-partner window.
  • Design-partner cohort opens today. A small founding cohort of IP firms — boutique and solo practitioners first — gets early-access pricing in exchange for structured feedback on the modules that matter most to their practice. The cohort is picked for fit, not for volume.
  • Honest launch state. PatentFlow has no paying customers at launch. It is opening a cohort today, not reporting adoption. The proof on offer is the architecture and the six-module platform — not invented traction.

The Velocity Process notes

What Claude Code handled. The full React + TypeScript + Vite + Tailwind + shadcn/ui single-page application across all six modules; the AWS Lambda + API Gateway v2 backend with its matters, AI-search, AI-drafting, AI-analysis, docketing, documents, filing, and USPTO-proxy handlers; the async job-and-worker pattern for long-running AI calls; the USPTO integration layer over the ODP, PatentsView, and PTAB APIs; the DynamoDB single-table data model and S3 document storage; the DOCX rendering for drafting output; the Cognito authentication; and the AWS CDK v2 infrastructure on the Silverback VelocityStack base class, deployed to a dedicated AWS account with the mandatory cost-attribution tags and the $500/mo budget alarm.

What required human judgement. The decision to launch the full six-module lifecycle as one platform rather than shipping a single module and calling it a wedge — the “five tools is not a strategy” thesis only holds if the lifecycle is actually integrated; the decision to route all inference through AWS Bedrock rather than a consumer API, because the patent-specific disclosure risk makes the enterprise envelope mandatory, not merely preferable; and the scoping discipline on litigation and international support — US prosecution and the litigation-support basics ship in the launch cohort, while deeper litigation-funding tooling and full EPO / JPO prosecution are roadmap, named honestly as roadmap rather than implied as shipped.

What the readiness pass caught. The Day 21 readiness audit caught two integrity problems on the launch surface, and both were pulled before go-live. First, the original landing page carried fabricated traction metrics — a stats bar claiming “200+ law firms” and “50,000+ patents processed” and a closing line about “hundreds of patent attorneys” already using the product. PatentFlow has no paying customers; it is opening a design-partner cohort today. The metrics were invented copy and were stripped — the same class of catch as the Day 14 GEOPress fabricated-”Trusted by 50+ firms” and the Day 15 Swole Labor fabricated-reviews. Second, the page claimed “SOC 2 Type II compliance”; SOC 2 is a roadmap milestone, not a current certification, and the claim was removed. The redesigned page leads with what is true — the six-module architecture, the Bedrock posture, the USPTO integration — and an honest design-partner CTA. The readiness-audit discipline catching fabrication before launch is itself part of the Velocity Process; the audit is published as gtm/DAY-21-READINESS-AUDIT.md.

What I’d do differently. Write the marketing surface last, against the verified product state, not first against the business-case vision. The fabricated metrics did not come from bad intent — they came from a landing page drafted ahead of the product’s actual adoption and never reconciled against it. The fix for the next venture is sequencing: the launch surface gets written after the readiness audit establishes what is true, not before.

What’s next this week

  • Day 21 (today): getpatentflow.com is live; the six-module platform is deployed on AWS; inference runs on Claude Sonnet 4.6 and Haiku 4.5 via AWS Bedrock; the USPTO integration reads ODP, PatentsView, and PTAB; the design-partner cohort opens.
  • Day 22 (Fri May 22): WhatAreWeWearing — the IRL Premium arc opens. TFTSL Week 3 wrap drops the same morning.
  • This week — punch list: resolve the launch-readiness items from gtm/DAY-21-READINESS-AUDIT.md Re-Audit #1 — the trial CTA wording, the credential scrub in the USPTO setup doc, founder-name consent, and an end-to-end verification that the Bedrock model id resolves on a real AI call.
  • Roadmap, named honestly as roadmap: deeper litigation-support and settlement / litigation-funding tooling; full EPO / JPO prosecution workflows; SOC 2 readiness; firm-specific workflow customization. None of these are claimed as shipped.

Want to talk

If you run an IP boutique or practice patent law, PatentFlow’s founding design-partner cohort is open today — book a demo at getpatentflow.com.

If you are a solo or small-firm practitioner priced out of $775-a-month single-purpose tools, the full-lifecycle platform is the cohort to be in.

If you are a regulated-practice operator weighing what enterprise-tier AI means for your firm, the Bedrock-routed architecture is a public reference point — the same posture CounselExpress runs on from Day 10.

PatentFlow reopens the Regulated Legal arc and lives inside the operator-validated-founder pattern the 30-day run is organized around. Live today at getpatentflow.com. Tomorrow: Day 22 — WhatAreWeWearing, the IRL Premium arc opens.

Live Product →