Autonomous Legal RAG Agent & Analytics Engine

An advanced Agentic RAG (Retrieval-Augmented Generation) system for constitutional and legal document reasoning using ReAct orchestration, dual-vector retrieval pipelines, ChromaDB, and Groq Llama-3.

The system supports:

• Semantic Retrieval
• Parent-Child Context Retrieval
• Dynamic PDF Upload Analysis
• Temporary In-Memory Vector Databases
• ReAct-based Agentic Reasoning
• Evaluation Metrics for RAG Performance

---

Features

Agentic ReAct Workflow

Implements iterative:

• Thought
• Action
• Observation
• Final Answer

reasoning loops for autonomous legal analysis.

Dual Retrieval Architecture

Two specialized retrieval pipelines:

Semantic Retrieval

Retrieves highly precise sentence-level legal fragments.

Parent-Child Retrieval

Retrieves broader contextual constitutional/legal sections for structural reasoning.

Dynamic PDF Upload Pipeline

Users can upload PDFs and query documents in real-time using:

• Ephemeral ChromaDB
• Temporary in-memory vector stores
• Runtime indexing

without polluting the permanent database.

Evaluation Framework

Tracks:

• Accuracy
• Faithfulness
• Relevance
• Failure Cases

for RAG quality analysis.

---

Tech Stack

• Python
• Streamlit
• Groq API
• Gemini API
• Llama-3
• ChromaDB
• Sentence Transformers
• LangChain
• PyPDF
• ReAct Agents

---

Project Architecture

User Query
      ↓
Agent Orchestrator (ReAct Loop)
     ↓
Tool Selection
 ┌───────────────┬────────────────┐
 │               │                │
Semantic Vault   Parent-Child Vault
 │               │
Precise Search   Contextual Search
 └───────────────┴────────────────┘
     ↓
Observation Injection
     ↓
Final Legal Synthesis

---

Folder Structure

project/
│
├── app.py
├── agent_orchestrator.py
├── rag_engine.py
├── rag_engine_v2.py
├── upload_pipeline.py
├── evaluation.ipynb
├── chroma_storage/
├── chroma_storage_v2/
├── requirements.txt
└── README.md

---

Installation

Clone Repository

git clone https://github.com/pranay0026/LegalRag.git cd LegalRag

Create Virtual Environment

python -m venv venv

Activate Environment

Windows

venv\\Scripts\\activate

Linux/Mac

source venv/bin/activate

---

Install Dependencies

pip install -r requirements.txt

---

Environment Variables

Create a .env file:

GROQ_API_KEY=your_key_here
GROQ_API_KEY3=your_key_here

---

Run Application

streamlit run app.py

---

Evaluation Metrics

Current benchmark results:

Metric	Score
Accuracy	4.00 / 5
Faithfulness	3.95 / 5
Relevance	4.00 / 5

---

Example Queries

Constitution Database Queries

• Powers of CMs under Article 162
• Fundamental Rights under Article 21
• Governor discretionary powers
• Emergency provisions in Constitution

Dynamic PDF Queries

• Summarize this agreement
• Identify legal risks
• Extract obligations from contract
• Explain termination clauses

---

Future Improvements

• Hybrid Search (BM25 + Vector Search)
• Reranking Pipelines
• Streaming Responses
• Conversational Memory
• FastAPI Backend
• Multi-document Retrieval

---

#Images

Author

Pranay Koppineedi

---

License

This project is intended for educational and research purposes.