Unpatent

Inspiration

Our inspiration to build Unpatent came from our own experience dealing with the exact problem we solve with Unpatent. We were creating our own startup but didn't know how related our product was to our competition and didn't have the budget to hire a patent lawyer, so we had to manually go through every single patent filed by our competitors and manually write down what we could and couldn't do.

What it does

Unpatent is an AI powered patent simplification tool built for solo inventors and early stage founders who can't afford a patent lawyer. Users upload patent PDFs, images, or pasted text, and Unpatent handles the rest. It extracts text from PDFs and scanned images using OCR, then simplifies dense legal patent language into a plain English breakdown covering the summary, simplified claims, diagram explanations, key terms, and risk areas. Results are visualized through interactive charts showing section breakdown, claim complexity, and protection focus. Users can also chat with a streaming AI patent assistant grounded in their uploaded documents. Everything is saved to a personal project workspace, with shareable 30-day public links and PDF export so founders can easily loop in advisors or investors. The goal is to give anyone the ability to understand what a patent actually covers without needing a law degree or a lawyer.

How we built it

Unpatent is built on Next.js 16 with the App Router, React 19, and TypeScript, deployed on Railway. For AI, it uses the OpenAI API gpt-4o-mini for fast structured extraction tasks and gpt-4o for richer reasoning and narrative simplification. Document processing relies on unpdf and pdfjs-dist for PDF text extraction, with tesseract.js handling OCR on scanned documents and images. Supabase powers the backend with Postgres, Auth, and Storage for user accounts, project workspaces, chat history, and shareable analysis tokens. The UI is built with Tailwind CSS 4, Framer Motion for animations, Recharts for data visualizations, and WebGL visual effects on the landing page.

Challenges we ran into

One of the first major challenges was scanned patent PDFs. Many patents are image-based with no selectable text, so we built a full OCR fallback using Tesseract.js with clear messaging when documents needed to be re-uploaded as images. Prompt engineering for legal language was equally tricky, patent claims are intentionally broad and ambiguous, and simplifying them accurately without hallucinating legal meaning required significant iteration. Responsible AI framing was a deliberate focus throughout, since getting things wrong could lead someone to infringe a patent or abandon a viable idea. Finally, supporting PDFs, images, and raw text across multiple simultaneous uploads required careful error handling at every step.

Accomplishments that we're proud of

The team built a tool that makes patent documents genuinely readable for non-lawyers, something that typically costs hundreds of dollars per hour with an attorney. The streaming chat assistant stays grounded in the actual uploaded patent content rather than hallucinating general patent law. Shareable workspaces with PDF export make the output practically usable in the real world for sharing with co-founders, advisors, and investors. The product also handles the full range of patent document formats, digital PDFs, scanned image PDFs, raw images, and pasted text, all within a single upload flow.

What we learned

Patent language is designed to obscure rather than communicate, and simplifying it without losing legal meaning turns out to be a harder AI problem than it first appears. Non-technical users, inventors, not lawyers, treat every piece of jargon as a reason to leave, which means plain language isn't a nice to have but the core product itself. OCR on real-world scanned documents is also messier than expected; government patent documents in particular can be low-resolution, rotated, or multi-column, all of which break naive extraction pipelines.

What's next for Unpatent

The roadmap includes competitor patent monitoring, which would watch for newly filed patents in a user's domain and alert them to relevant publications, turning Unpatent from a one-time lookup into ongoing IP intelligence. Side by side comparison would let users upload multiple patents and highlight exactly where they overlap or diverge. Claim drafting assistance would help users draft their own claims in plain language after understanding what's already patented, before handing off to a lawyer. Finally, team workspaces would allow co-founders and advisors to collaborate on the same project without sharing login credentials.

Built With

• framer-motion
• next.js
• openai-api
• pdfjs-dist
• railway
• react-19
• recharts
• supabase
• tailwind-css-4
• tesseract.js
• typescript
• unpdf