PeroAI is an AI governance layer that ensures safe, compliant, and traceable prompt interactions with generative AI models. Built with AWS Bedrock, Amazon Polly, and a multi-agent architecture powered by the Model Context Protocol (MCP), PeroAI is designed for enterprise-grade GenAI oversight.

Organizations deploying GenAI face risks from:

• Unsafe or toxic prompt input
• Leaked sensitive data via AI outputs
• Lack of control, traceability, or policy enforcement

There's no standard way to moderate, audit, and govern AI model usage across tools like ChatGPT, Claude, or Bedrock.

PeroAI acts as a secure middle layer that intercepts, evaluates, and governs prompts using:

• ✅ Multi-agent Governance via MCP
• ✅ Prompt moderation and rewriting
• ✅ AI response generation via AWS Bedrock
• ✅ Speech synthesis via Amazon Polly
• ✅ Audit logging via DynamoDB
• ✅ 3D avatar interface with viseme lip sync

• Frontend: Next.js + React + Three.js (3D avatar)

• Backend: FastAPI

• AI Platform: AWS Bedrock (Claude v2)

• Voice: Amazon Polly (speech + viseme)

• Storage: DynamoDB (logs)

• MCP Agents:

  • PolicyAgent → Check compliance
  • RewriteAgent → Rephrase if blocked
  • ResponseAgent → Generate safe response
  • VoiceAgent → Convert to voice
  • TraceAgent → Record everything

A short 2–3 minute video walking through prompt input → governance → spoken response with avatar.

# Frontend
cd app
npm install
npm run dev

# Backend
cd backend
pip install -r requirements.txt
uvicorn main:app --reload

Add the following env vars:

OPENAI_API_KEY=xxx
AWS_ACCESS_KEY_ID=xxx
AWS_SECRET_ACCESS_KEY=xxx
AWS_REGION=us-east-1

• 🔒 Strong alignment with governance & safety theme
• 🧠 Deep use of AWS-native tech: Bedrock + Polly
• 🎨 Engaging UX via 3D Avatar with lip sync
• 📊 Transparent audit trails for enterprises
• 🧱 MCP = Composable, flexible AI control system

Built solo by Mungunshagai, with support from ChatGPT + AWS open source libraries.

• GitHub Repo: github.com/mongonsh/pero-ai
• Live Demo: pero-ai-sandy.vercel.app

MIT