Landing Page
Quick Summary
Deep Review

ContractPilot - AI Contract Review Anyone Can Understand

Inspiration

Most people sign contracts they don’t fully understand because legal help is expensive ($200–500/hr). One hidden clause (non-compete, IP transfer, auto-renewal) can cost thousands. ContractPilot gives you a fast, plain-English risk review for the price of a coffee.

What it does

Upload a contract (PDF, Word, or scanned paper) and get:

Overall Risk Score (0–100) with an animated gauge
Risk breakdown: Financial, Compliance, Operational, Reputational (computed from clause-level data via Dedalus tools)
Clause-by-clause explanations: “What it means,” “Watch out,” “Suggested change” (no legal jargon)
Deep Review Mode: side-by-side PDF viewer with color-coded highlights; hover to see analysis, click to chat
Action checklist + key dates timeline extracted directly from the document
Downloadable PDF report
Real-time progress as clauses stream in
Dark mode
Pricing: first review free, then 5 reviews for $2.99

How we built it (high level)

Frontend

Next.js 16 (App Router, Turbopack)
Tailwind CSS v4 and Framer Motion for UI and animations
react-pdf for interactive, highlighted PDF viewing
jsPDF for downloadable reports
Convex for real-time state and credit-based paywall
Convex Auth (Google OAuth) for authentication

Backend

Python + FastAPI for orchestration
PyMuPDF for text extraction and text-to-page coordinate mapping
python-docx for Word document ingestion
Tesseract OCR (local, no cloud API) for scanned documents with word-level bounding boxes

AI & Intelligence Layer

Clause extraction pipeline: regex sectioning → sub-clause splitting → intelligent filtering
K2 Think (kimi-k2-instruct via Vultr Serverless Inference) for parallel clause analysis (6 concurrent)
Vultr RAG with llama-3.3-70b for grounded legal retrieval
Legal data: CUAD (500+ expert-annotated contracts, 41 clause types) + Legal Clauses dataset (21K+ clauses)

Agentic Orchestration & Chat

Dedalus ADK (Python) as the primary agent framework
Native Dedalus tools: compute_risk_breakdown, find_key_dates, search_legal_knowledge_base
MCP servers: Brave Search (legal web context) and Exa (academic/legal research)
Dedalus Auth (DAuth) secures MCP credentials (no third-party keys stored in app code)

Realtime UX

Clause-level results stream through Convex so the UI updates live
Deep Review Mode transforms static PDFs into interactive, color-coded documents with hover-to-reveal analysis and click-to-chat

Challenges

Handling session-based auth identifiers in Convex without breaking ownership checks
Precisely highlighting clauses in PDFs (PyMuPDF search with OCR fallback)
Balancing speed vs. depth: parallel deterministic analysis for clauses, agentic reasoning for synthesis
Forcing plain-English output instead of legal jargon through prompt iteration