🚀 Intelligent Document Assistant
A smart, AI-powered enterprise document assistant built with Gemini API, LangChain, and Flask, designed to analyze contracts, policies, legal documents, and internal reports with ease.
This assistant can:
🔍 Extract clauses
📑 Summarize documents
🤖 Answer questions using RAG
📤 Upload & ingest PDFs, DOCX, TXT
🧠 Maintain searchable history
🕸️ Provide LangSmith export
📈 Offer auto-escalation for high-risk clauses
✨ Features 🔹 1. Intelligent Q&A
Ask any question from uploaded files — the system retrieves the most relevant chunks and generates human-like answers using Google Gemini.
🔹 2. Document Upload & Ingestion
Supports:
PDF (with fallback to OCR for scanned files)
DOCX
TXT
Automatic chunking + embedding using SentenceTransformers.
🔹 3. Risk Analysis
Identifies:
Liability clauses
Indemnities
Termination conditions
Non-compete and confidentiality risks
High-risk results are auto-escalated.
🔹 4. Conversation History
Every query is preserved with:
Question
Answer
Context used
Sources
Risk score
History is searchable & replayable.
🔹 5. LangSmith Export
Export your entire dataset as JSON for evaluation, monitoring, or training.
🔹 6. Clean UI
Single-page interface with:
Ask panel
Upload & ingest integrated inside Ask
Scrollable results
Right-side History panel
🏗️ Tech Stack Component Technology Backend Python, Flask LLM Google Gemini (2.5 Flash / 1.5 Flash fallback) Embeddings SentenceTransformers MiniLM Retrieval Custom retriever + vector store Frontend HTML5, JavaScript, Marked.js (Markdown rendering) Storage SQLite (history), filesystem (uploads) Dev Tools LangChain, LangGraph, LangSmith 📂 Project Structure DOC_assistant/ │── app.py │── requirements.txt │── README.md │── .gitignore │── templates/ │ └── index.html │── static/ │ └── style.css │── services/ │ ├── ingest.py │ ├── llm_client.py │ ├── retrieval.py │ ├── risk.py │ └── db.py │── data/ │ ├── uploads/ │ └── history.db (auto-created) └── tests/
⚙️ Installation
-
Clone repo git clone https://github.com/Akash7367/DOC_assistant.git cd DOC_assistant
-
Create virtual env python -m venv myenv myenv\Scripts\activate
-
Install dependencies pip install -r requirements.txt
-
Create .env file
(Do NOT commit this.)
GEMINI_API_KEY=your_api_key_here GEMINI_MODEL=gemini-1.5-flash SECRET_KEY=your_flask_secret_here
Optional OCR support for scanned PDFs:
pip install pillow pytesseract pdf2image
Open in browser:
🧠 How It Works (RAG Pipeline)
1️⃣ Upload Document → Extract text → Chunk → Embed → Store
2️⃣ Ask Question → Retriever selects best chunks → Sent to Gemini
3️⃣ LLM Generates → Answer → Markdown formatting → Risk report → Sources
4️⃣ History Saved → SQLite entry created → Can be replayed anytime
📊 Architecture Diagram (Add your PNG here later) [User] → [Frontend] → [Flask Backend] → [Retriever] → [Embeddings DB] ↓ [Gemini LLM] ↓ [Final Answer]
📸 Screenshots (You can upload images in GitHub)
Add like:
🧪 Testing pytest
📤 Export to LangSmith
Go to:
/export/langsmith
Downloads langsmith_export.json.
🛡️ .env Protection
Your repo already includes a strong .gitignore:
✔ .env excluded ✔ myenv/ excluded ✔ Sensitive files protected
🤝 Contributing
Pull requests are welcome! For major changes, open an issue first.
📜 License
MIT License (or specify)
🎉 Done!
Your README is now professional, clean, and GitHub-ready.
If you want:
A logo for your project
Architecture diagram PNG
GIF demo
More badges (build passing, stars, issues, license) Just tell me — I can generate all of them.