ClauseIQ

Project Overview ### Purpose ClauseIQ is a demonstration of enterprise-grade AI engineering, built to be deployed as a portfolio project on freelance platforms and as a foundation for a potential SaaS product. ### Scope **In Scope:** - PDF contract upload and processing - AI-powered clause extraction, classification, and risk scoring - Question-answering with semantic search and LLM grounding - User authentication and multi-organization data isolation - Production-grade error handling, logging, and rate limiting - React frontend with real-time UI updates - Docker containerization for easy deployment **Out of Scope (for now):** - Mobile apps (but the web app is responsive) - Real-time collaboration (single-user per document) - Advanced CMS or document versioning - Integration with legal case management systems (future roadmap) - Audit trails or compliance logging (beyond basic auth logs) ### Main Features | Feature | What It Does | Why It Matters | |---|---|---| | **RAG Pipeline** | Upload PDF → parse text → split into chunks → embed with Gemini/OpenAI → store in pgvector → retrieve top-K matching chunks based on semantic similarity → feed to LLM for grounded answer | Answers are cited — the LLM tells you exactly which clause it's basing the answer on | | **Clause Classification** | Each chunk gets classified as termination, liability, payment, etc. | Users can filter or focus on specific clause types | | **Risk Scoring** | Each clause gets a risk level (high/medium/low) based on deviation from standard language | Paralegals and lawyers review the high-risk items first instead of reading 50 pages | | **Chat Interface** | Ask natural language questions; get answers with source excerpts | Non-technical users can find contract info without manual search | | **Multi-Org Auth** | Sign up creates an org + user; every document is org-scoped | Each customer's contracts stay private; org A cannot read org B's data | | **Async Analysis** | Upload doesn't block on AI processing; user can ask questions while analysis is running | Better UX than "please wait 60 seconds" | | **Error Recovery** | Rate limiting, validation, timeout handling, graceful LLM failures | Won't crash on edge cases; won't run up huge bills | ### Modules ``` Backend (FastAPI + PostgreSQL + pgvector) ├── Auth (JWT + password hashing) ├── RAG Pipeline (parse → chunk → embed → retrieve → answer) ├── Risk Analysis (classify + score clauses) ├── Logging & Error Handling ├── Rate Limiting └── API Endpoints (24 routes) Frontend (React 18 + Tailwind-equivalent CSS) ├── Auth Pages (signup/login) ├── Document Upload ├── Risk Dashboard ├── Clause Browser ├── Chat Q&A └── Sidebar Navigation ``` ### User Roles | Role | Who | Permissions | |---|---|---| | **Org Admin** | First user created at signup | Full access to all org's documents, but cannot see other orgs' data | | **Org Member** | Additional users added later (infrastructure exists, but no UI yet) | Same as admin for now | | **Anonymous** | No login | Can sign up or log in | ### Functionalities 1. **Authentication**: Signup creates org + user; login returns JWT token 2. **Document Management**: Upload, list, retrieve, delete documents 3. **Processing**: Auto-extract text, chunk, embed, store embeddings 4. **Analysis**: Scan all chunks for clause type + risk level + explanation 5. **Retrieval**: Semantic search across document corpus 6. **Q&A**: Answer questions with AI, cite sources 7. **Logging**: Every request logged with ID, method, path, status, duration 8. **Rate Limiting**: Per-org or per-IP throttling on expensive endpoints 9. **Error Handling**: Clean, non-leaking error messages ### Business Goals 1. **Demonstrate RAG competency**: Show that you understand embeddings, similarity search, and grounded generation 2. **Prove production readiness**: Logging, error handling, auth, and deployment work out of the box 3. **Enable fast sales cycles**: "Here's a working demo; here's the code on GitHub" — no more explanation needed 4. **Support billable services**: Model to charge per-contract-analyzed or per-month subscription on Upwork ### Project Vision From a prototype that works locally to a deployed, revenue-generating service: ClauseIQ proves that modern AI engineering isn't just about clever prompts — it's about systems thinking, reliability, and user experience. ---