I'm building an AI tool that helps non-lawyers and busy procurement/legal teams quickly review vendor/client contracts, NDAs, employment agreements, etc. — without uploading sensitive data to the cloud (offline/local-first option) or replacing lawyers.
Background: As someone who's wasted days manually hunting for risky clauses, vague terms, hidden overrides in amendments, or unfair liability language in vendor deals, I decided to prototype this after seeing how much time/money gets burned on basic reviews.
What it does right now (early MVP/beta): - Upload PDF/Word/plain text contract - Scans for common risks: indemnity caps missing, auto-renewals, one-sided termination, IP ownership traps, liability exceeding fees, non-compete overreach, etc. - Flags issues with plain-English explanations + confidence score - Suggests safer alternative clauses (based on standard templates/best practices) - Basic redlining/highlighting output (exportable) - Offline mode using local models (no data leaves your machine)
Tech stack (simple & transparent): - Frontend: React + Tailwind - Backend: Python + fine-tuned open models (e.g., Llama-3 or similar legal-tuned variants) + some rule-based checks for accuracy - No cloud LLM calls in core flow (privacy focus); optional Grok/Claude integration for deeper suggestions - Processes docs locally via Ollama or similar
Current status: - Tested on ~50 real-ish contracts (NDAs, SaaS agreements, freelance templates) - Average time: 2-5 minutes vs. hours/days manual - ~75-85% of obvious risks caught (still misses nuanced stuff — not lawyer-grade yet) - Free beta, no signup required (just drag & drop on the demo page)
I'm looking for brutal feedback, especially from: - In-house counsel/procurement folks: What clauses cause you the most pain/headaches? - Developers/freelancers/small biz owners: Would you trust this for quick scans before signing vendor deals? - Anyone who's used Spellbook/LegalFly/Ironclad: How does this compare? What gaps do you see? - Trust/accuracy concerns: Hallucinations, false positives, liability disclaimers?
Happy to share more on training data approach, offline setup, or why I focused on negotiation basics vs. full lifecycle.
Thanks for any thoughts — this is day-early, so roast away!