AI Document Intelligence Platform

Production-ready PDF analysis powered by RAG — upload, process, search, and extract intelligent insights from PDF documents.

Architecture

Frontend (SPA)  ──REST──▶  FastAPI  ──▶  Services  ──▶  ChromaDB + SQLite
                                          │
                                    PyMuPDF · Tesseract
                                    pdfplumber · sentence-transformers
                                    OpenAI GPT · InsightsService

Layer	Tech	Purpose
Frontend	Vanilla HTML/CSS/JS	Upload UI, document manager, insights viewer, chat
API	FastAPI + Uvicorn	REST endpoints, SSE streaming, middleware
PDF Engine	PyMuPDF, pytesseract, pdfplumber	Text extraction, OCR, tables
Chunking	Recursive splitter	Semantic text segmentation
Embeddings	all-MiniLM-L6-v2	384-dim dense vectors
Vector DB	ChromaDB (persistent)	Similarity search
Metadata	SQLite (aiosqlite)	Document state, insight caching
LLM	OpenAI GPT-4o-mini	Chat, insight generation
Export	reportlab, python-docx, openpyxl	PDF, Word, Excel export

Features

Document Intelligence

• Executive Summary — concise overview of document content
• Key Insights — top findings and takeaways
• Action Items — extracted tasks and next steps
• Risks — identified warnings and compliance concerns
• Recommendations — proposals from the document
• KPIs & Metrics — extracted performance indicators
• Important Dates — deadlines and milestones
• Named Entities — People, Organizations, Locations
• Financial Metrics — revenue, costs, budget data
• Table Insights — trends from tabular data
• Multi-document Comparison — cross-document analysis

Document Management

• Upload multiple PDF files (drag & drop)
• View uploaded documents with processing status
• Delete documents and all associated data
• Duplicate detection — SHA-256 hash prevents re-indexing identical files

RAG Chat

• Semantic search over indexed documents
• Streaming SSE responses
• Conversation history
• Citations with document name, page number, and source text

Export

• Export insights as PDF, Word (DOCX), or Excel (XLSX)
• Export chat conversation history as TXT

Production Features

• Structured logging with request IDs
• Global exception handling with error mapping
• Health checks (metadata store, vector store, embedding model, LLM)
• Docker multi-stage build with non-root user
• Unit tests and integration tests
• CORS middleware, request timing

Quick Start

Option 1: Docker (Recommended)

cp .env.example .env
# Edit .env to add your OPENAI_API_KEY
cd docker
docker compose up --build

Open http://localhost:8000 in your browser.

Option 2: Bare-Metal

Prerequisites:

• Python 3.11+
• (Optional) Tesseract OCR for scanned PDFs
• (Optional) Poppler for pdf2image

# 1. Create virtual environment
cd backend
python -m venv .venv

# Windows
.venv\Scripts\activate

# macOS/Linux
source .venv/bin/activate

# 2. Install dependencies
pip install -r requirements.txt

# 3. Configure
cp ../.env.example ../.env
# Edit .env to add your OPENAI_API_KEY

# 4. Run
uvicorn app.main:app --host 0.0.0.0 --port 8000 --reload

Open http://localhost:8000 in your browser.

API Reference

Core Endpoints

Method	Endpoint	Description
GET	/api/health	System health check
POST	/api/documents/upload	Upload PDF files (multipart)
GET	/api/documents	List all documents
GET	/api/documents/{id}	Get document details
DELETE	/api/documents/{id}	Delete document
GET	/api/documents/{id}/chunks	Inspect document chunks

Intelligence Endpoints

Method	Endpoint	Description
GET	/api/documents/{id}/insights/{type}	Get/generate insight
POST	/api/documents/compare	Compare multiple documents

Insight types: executive_summary, key_insights, action_items, risks, recommendations, kpi_metrics, important_dates, named_entities, financial_metrics, table_insights

Chat & Search

Method	Endpoint	Description
POST	/api/chat	Chat with documents (SSE stream)
GET	/api/search?query=...	Semantic search
GET	/api/conversations	List conversations
POST	/api/conversations	Create conversation
GET	/api/conversations/{id}	Get conversation history

Export

Method	Endpoint	Description
GET	/api/export/insights/{doc_id}/{format}	Export insights (pdf/word/excel)
GET	/api/export/chat/{conv_id}/txt	Export chat history

Upload Example

curl -X POST http://localhost:8000/api/documents/upload \
  -F "files=@document.pdf"

Generate Insight Example

curl http://localhost:8000/api/documents/{doc_id}/insights/executive_summary

Running Tests

cd backend
pytest tests/ -v

Configuration

All settings are configurable via environment variables (see .env.example):

Variable	Default	Description
OPENAI_API_KEY	—	OpenAI API key (required for chat/insights)
LLM_MODEL	gpt-4o-mini	LLM model for generation
MAX_FILE_SIZE_MB	50	Maximum upload size
CHUNK_SIZE	1000	Target chunk size (chars)
CHUNK_OVERLAP	200	Overlap between chunks
EMBEDDING_MODEL	all-MiniLM-L6-v2	Sentence-transformer model
TOP_K_RETRIEVAL	5	Number of chunks to retrieve
SIMILARITY_THRESHOLD	0.5	Distance threshold for filtering
OCR_ENABLED	true	Enable Tesseract OCR
LOG_LEVEL	INFO	Logging verbosity

Project Structure

├── backend/app/
│   ├── core/             # Domain models, interfaces, exceptions
│   ├── services/         # PDF extraction, chunking, embeddings, insights, chat
│   │   ├── insights_service.py   # Document intelligence generation
│   │   ├── chat_service.py       # RAG chat orchestration
│   │   └── ...
│   ├── storage/          # ChromaDB + SQLite implementations
│   ├── api/routes/       # REST endpoints
│   │   ├── documents.py  # Upload, list, delete, insights
│   │   ├── chat.py       # Chat + conversations
│   │   ├── export.py     # PDF/Word/Excel/TXT export
│   │   └── health.py     # Health checks
│   ├── config.py         # Environment-based configuration
│   ├── dependencies.py   # Dependency injection
│   └── main.py           # FastAPI app factory
├── backend/tests/        # Unit + integration tests
├── frontend/             # Single-page UI
├── docker/               # Containerization
└── data/                 # Runtime data (gitignored)

License

MIT