SentinelGov-AI

Here is the revised About the Project section in Markdown format with all emojis removed, keeping a formal and professional tone:

About the Project: SentinelGov-AI

Inspiration

With the rapid adoption of Generative AI in sensitive domains like healthcare, finance, and law, the lack of governance, transparency, and ethical safeguards became a central concern. We were inspired to build a solution that not only responds to user queries but also audits, interprets laws, preserves privacy, and ensures bias-free decisions.

The GenAI Hackathon by Impetus and AWS offered the perfect theme: “Multi-agent governance system for ethical and compliant AI”, and we knew instantly this was the challenge we wanted to tackle.

What We Built

SentinelGov-AI is a cloud-based multi-agent governance chatbot tailored for the healthcare industry. When a patient submits a query (e.g., "Delete my cancer report" or "Book an appointment under GDPR rules"), our AI agents collaborate behind the scenes to ensure the request is handled lawfully and ethically.

It’s more than a chatbot — it’s a real-time governance framework with:

Compliance Agent (HIPAA, GDPR, EU AI Act)
Privacy Agent (Pseudonymization and risk detection)
Ethics Agent (Bias and fairness interpretation)
Policy Agent (Deadline enforcement, DSA, SLA, etc.)
Audit Agent (JSON-based traceable logging)

Technical Architecture

Our solution includes:

FastAPI backend (core logic and orchestration)
React + Vite frontend (user interface and live logs)
Claude 3.5 via Amazon Bedrock (LLM reasoning)
Amazon S3 and optional blockchain (audit log persistence)
Custom rules engine in JSON (policy_store.json)
TTS generation pipeline (for narrated video demo)

Built With

Languages: Python, JavaScript
Frameworks: FastAPI, React
Cloud Services: AWS Bedrock, EC2, IAM, Lambda, S3
DevOps Tools: GitHub, pyenv, virtualenv
Other: Pseudonymization, audit trace logging, Claude prompts

What We Learned

How to integrate LLMs with structured multi-agent workflows
Building auditable GenAI systems in regulated sectors
Deepened our skills in AWS Bedrock, Claude 3.5, and TTS narration
Learned how to design modular, legally aligned AI agents with explainability

Challenges Faced

Setting up AWS Bedrock and Claude 3.5 with secure region and credentials
Handling long prompts and decoding errors in TTS pipelines
Designing agent modularity while maintaining shared context
Ensuring fast inference and live log display under async execution

Final Thoughts

We’re proud to have built a GenAI system that goes beyond generation — it governs. SentinelGov-AI stands as a reference for what responsible AI should look like in critical domains.

Thank you, Impetus and AWS, for empowering us with this incredible platform.

Let me know if you'd like this saved to a .md file or added to your GitHub repo structure. tinelGov-AI