FindLaw by Claude

Inspiration

Legal advice is one of the few professional services where the people who need it most can afford it least. 77% of Americans facing a legal problem never consult a lawyer — not because they don't need help, but because a single consultation can cost $500.

We wanted to change that. Not by replacing lawyers, but by giving everyone access to the same foundation lawyers use: legal precedent.

What It Does

FindLaw is a conversational AI legal assistant that works like a good lawyer would in an initial consultation. Instead of immediately giving a generic answer, it:

Asks targeted follow-up questions to understand your specific situation
Tracks a confidence score — it won't give advice until it's confident it has enough detail
Searches across 7 live legal databases — national and international
Returns the actual court cases that apply to your situation, with plain-English summaries and links to the full rulings

The result: real legal guidance grounded entirely in precedent, not guesswork.

How We Built It

AI backbone: Claude (Anthropic) — handles the conversational flow, confidence scoring, and case reasoning
Legal databases: CourtListener, Harvard Caselaw Access Project, Oyez (SCOTUS), HUDOC (European Court of Human Rights), CURIA (EU Court of Justice), GovInfo, and the Federal Register
Frontend: Vanilla HTML/CSS/JS — fast to build, easy to demo
Backend: Node.js server proxying requests to the Anthropic API

The system prompts Claude to gather jurisdictional context first — country, and state if the user is in the US — before searching the relevant databases.

Challenges We Faced

Data sourcing was harder than expected. Most legal databases require registration, have CAPTCHA walls, or throttle unauthenticated requests. We had to identify which APIs were truly open and reliable under time pressure.

Scoping the AI behaviour. Getting Claude to ask just enough questions — not too few (vague advice) and not too many (frustrating UX) — required careful prompt engineering around the confidence scoring mechanism.

Time. 90 minutes is not a lot of time to build, test, and pitch a legal AI.

Accomplishments We're Proud Of

A working confidence scoring system that mimics how a real lawyer gathers facts before giving advice
Live integration with multiple free legal databases across US and international jurisdictions
A clean, professional UI that feels trustworthy — important for a legal product
Built and pitched in 90 minutes

What We Learned

Scoping fast is a superpower. We had bigger ideas — video ingestion, document upload, multi-jurisdiction comparison — and deliberately cut them to ship something that works and demos well. A focused, polished prototype beats an ambitious, broken one every time.

We also learned that legal data is surprisingly accessible when you know where to look. Millions of court decisions are public record and available via free APIs — the barrier isn't the data, it's knowing it exists.

What's Next for FindLaw

Document upload — let users submit contracts, letters, or court notices for case-matched analysis
Expanded international coverage — CanLII (Canada), BAILII (UK), and more
A citator feature showing whether a cited case is still good law
Voice input for accessibility

The foundation is built. The precedent is set. ⚖️