🧠 DocBrain — Universal Business Document Intelligence API

A production-grade RAG (Retrieval-Augmented Generation) API that lets any business upload documents and instantly query them with natural language. Built with FastAPI, ChromaDB, and Claude.

Why This Exists

Every business has the same problem: critical knowledge locked in PDFs, contracts, reports, and SOPs that nobody can find or query quickly. DocBrain solves this with a clean REST API any team can integrate in minutes.

Use cases across industries:

Industry	Use Case
Legal	Query contracts, extract obligations and deadlines
Finance	Summarize quarterly reports, extract financial figures
HR	Search policy manuals, answer employee questions
Healthcare	Parse clinical SOPs, extract protocols
Sales	Query RFPs, compare vendor proposals
Real Estate	Analyze lease agreements, flag risk clauses

---

Architecture

Client Request
     │
     ▼
FastAPI (app/main.py)
     │
     ├── /collections  ──► Collections Router
     ├── /documents    ──► Ingestion Pipeline
     │                         ├── Text Extraction (PDF/TXT/MD/CSV)
     │                         ├── Sentence-aware Chunking
     │                         └── ChromaDB Vector Storage
     └── /query        ──► LLM Layer (app/llm.py)
                               ├── Semantic Search (ChromaDB)
                               └── Claude Generation (Anthropic API)

RAG Pipeline:

1. Ingest — documents are parsed, chunked (800 tokens, 150 overlap), embedded via sentence-transformers, stored in ChromaDB
2. Retrieve — incoming questions trigger cosine-similarity search across stored chunks
3. Generate — top-k chunks passed as context to Claude with grounding instructions
4. Cite — response includes source citations with relevance scores