🔍 RAG-Based Policy Agent

Enterprise-Grade Policy Intelligence Platform Powered by Retrieval-Augmented Generation (RAG)

Transform fragmented regulatory PDFs into a searchable, intelligent, context-grounded policy assistant using semantic embeddings, FAISS retrieval, and Google Gemini reasoning.

---

📘 Overview

RAG-Based Policy Agent is a production-ready platform that converts PDF-based policies, compliance rules, and regulatory frameworks into a real-time policy intelligence engine.

It enables policy teams to instantly search, retrieve, and reason over internal documents — with grounded, traceable, and compliant answers powered by RAG.

Designed for:

• 🛡 Compliance & Audit
• ⚖ Legal & Governance
• 🏢 Corporate Policy Teams
• 🔍 Internal Risk & Regulatory Operations

---

🧠 Why This Product Matters

Organizations face challenges with:

• Thousands of pages of policy documents
• Manual search through static PDFs
• Slow compliance decision-making
• Risk of misinterpretation or oversight

This platform provides:

✓ Real-time policy intelligence
✓ Explainable, context-grounded LLM answers
✓ High-precision document extraction
✓ Enterprise-level consistency & reliability

---

🗂 Project Structure

srujanrana07-rag-based-policy-agent/
│
├── README.md                 
├── LICENSE
│
├── app.py                    # Flask REST API server
├── main.py                   # Core RAG workflow runner
├── deploy.sh                 # Deployment script
│
├── document_loader.py        # PDF/table loader & preprocessing
├── vectorizer.py             # Sentence Transformer embeddings
├── retriever.py              # FAISS vector store operations
├── gpt_client.py             # Gemini LLM wrapper
├── submitter.py              # Batch inference + evaluation
│
├── answers_output.json       # Output results
├── questions.json            # Test questions dataset
├── temp.json                 # Intermediate processing store
│
├── requirements.txt
└── log.txt

---

🏗 System Architecture

---

🚀 Key Features

1️⃣ Intelligent PDF Ingestion

• High-fidelity extraction via GhostScript
• Accurate table detection through Camelot-py
• Clean preprocessing for long-form policies

2️⃣ Embedding & Vectorization

• Sentence Transformer embeddings
• Chunking + semantic scoring
• Stored and indexed for fast retrieval

3️⃣ FAISS-Powered Semantic Search

• Millisecond-level semantic retrieval
• Top-k contextual selection
• Deterministic policies for safe RAG

4️⃣ Context-Grounded LLM Reasoning

• Gemini LLM integrated through gpt_client.py
• Strict grounding in FAISS-retrieved context
• No hallucinations → verified policy answers

5️⃣ Modular, Production-Ready REST API

• Upload documents
• Extract structured content
• Ask policy questions
• Plug into dashboards, portals, BI tools

---

⚙ Installation

Clone Repository

git clone https://github.com/srujanrana07/rag-based-policy-agent
cd rag-based-policy-agent

Create Virtual Environment

python -m venv venv
source venv/bin/activate
# Windows: venv\Scripts\activate

Install Dependencies

pip install -r requirements.txt

Configure Environment

Create a .env:

GEMINI_API_KEY=your_api_key

---

▶ Usage

Start API Server

python app.py

Run End-to-End RAG Pipeline

python main.py

Run Batch Q&A Evaluation

python submitter.py

---

🔌 API Endpoints

Method	Endpoint	Description
POST	/upload	Upload a policy PDF
GET	/extract	Extract text + tables
POST	/query	Ask a question using RAG

---

🔐 Security & Reliability

• Sanitized PDF ingestion
• Deterministic FAISS retrieval
• Strictly grounded LLM answers
• Full logging in log.txt
• Cloud-ready and container-friendly

---

🐳 Deployment

Docker

docker build -t rag-policy-agent .
docker run -p 5000:5000 rag-policy-agent

Linux (Auto Setup)

./deploy.sh

Supports deployment on:

• AWS (ECS, EC2, Lambda)
• GCP (Cloud Run)
• Azure App Service
• Railway / Render

---

🤝 Contributing

Pull requests and improvements are welcome.

---

📄 License

MIT License — free for personal and commercial use.