SentinelRAG: Private, Secure T&C and Contract QA Assistant (Discovery/Learning Prototype)

Overview

SentinelRAG is a self-hosted prototype designed to help you make sense of the legal documents you’ve signed — from Terms & Conditions and NDAs to privacy policies and contracts — simply by asking questions in plain English.

💡 The idea for this project was inspired by a real experience:

When I bought a home, I had to sign a large number of legal documents for the first time — and it made me realize how little visibility I had into what I was actually agreeing to. That experience sparked the idea for a system that would let anyone securely upload their own contracts and later ask meaningful questions about them — without needing to dig through legal jargon.

🔍 What SentinelRAG does

SentinelRAG lets you:

• Upload contracts, policies, or agreements
• Ask natural-language questions like:
  • “Can they share my data?”
  • “Am I locked into a renewal?”
  • “Who owns the intellectual property?”
• Receive fast, relevant answers that help you understand what you’ve agreed to — without reading every clause manually.

🔐 Why it matters

Legal documents shouldn’t be a black box. SentinelRAG aims to bring clarity, security, and control to your agreements, powered by Retrieval-Augmented Generation (RAG) and wrapped in best practices for DevOps and AI infrastructure.

This is an ongoing learning prototype — built to explore how GenAI, when combined with secure infrastructure, can make legal understanding more human-friendly and trustworthy.

Key Practices Demonstrated

• TLS/mTLS encryption and hardened Docker containers
• Observability via Galileo GenAI SDK (or Prometheus + Grafana fallback)
• Kubernetes-based CI/CD with Helm and GitHub Actions
• Modular, API-first design using FastAPI and LangChain
• Uses hardened infrastructure to keep sensitive data protected

🛡️ This is a prototype, not yet intended for production use. It is being developed to explore infrastructure, security, and reliability practices relevant to secure GenAI systems.

Stack

• Backend: Python, FastAPI, LangChain
• Retrieval: FAISS (persistent disk storage) (For now)
• Embeddings: HuggingFace (MiniLM-L6-v2)
• LLM: Mistral via Ollama (local inference using Ollama)
• Security: SHA256 deduplication, TLS/mTLS planned
• Deployment: Docker → K8s (via Helm)
• Observability: Galileo SDK (planned) or Prometheus + Grafana (fallback)
• DevOps: GitHub Actions (planned)

Status

• First commit: June 7, 2025
• MVP target: June 12, 2025
• Actively being developed — structure and pipeline under construction

Planned Structure

SentinelRAG/ SentinelRAG/

• backend/ # FastAPI app + LangChain pipeline
  • main.py # REST API (upload, ask, delete vectordb)
  • rag_pipeline.py
  • requirements.txt
• vectorstores/ # Persistent FAISS index + hash DB
• security/ # TLS, hardening configs (planned)
• k8s/ # Helm charts and manifests (planned)
• observability/ # Monitoring and tracing setup (planned)
• frontend/ # Streamlit UI (optional)
• README.md

UPDATES:

• Omitting security, untill the end