Transparens AI

Transparent AI answers with verifiable evidence.

Transparens AI is a system that provides AI-generated answers with complete transparency and source attribution. Unlike traditional AI systems that act as "black boxes," Transparens AI separates the research and answer generation processes, providing users with:

• Structured answers broken into clear, logical blocks
• Direct source citations for every claim
• Evidence graphs that visualize the relationship between questions, answers, and sources
• Full transparency into how answers are constructed

System Architecture

Overview

User Question
    ↓
Research Agent (Exa) → Retrieves relevant sources
    ↓
Answer Agent (LLM) → Generates answer ONLY using those sources
    ↓
Evidence Graph Builder → Links question → answer blocks → sources
    ↓
JSON Response → Frontend visualization

Key Design Principles

1. Separation of Concerns: Research and answer generation are separate, independent processes
2. Source Constraint: The LLM can ONLY use information from retrieved sources
3. Structured Output: Answers are broken into blocks (paragraphs/bullets) with explicit source citations
4. Transparency: Every claim is traceable to its source via the evidence graph

Repository Structure

MadHacks2025/
├── backend/              # Node.js/TypeScript API server
│   ├── src/
│   │   ├── index.ts              # Express server
│   │   ├── routes/               # API endpoints
│   │   ├── services/             # Core business logic
│   │   │   ├── researchAgent.ts  # Exa-based RAG
│   │   │   ├── answerAgent.ts    # LLM answer generation
│   │   │   └── evidenceGraph.ts  # Graph construction
│   │   ├── types/                # TypeScript definitions
│   │   └── config/               # Environment config
│   └── README.md
│
└── frontend/             # (Empty - future React/Viz app)
    └── .gitkeep

Tech Stack

Backend

• Language: TypeScript
• Runtime: Node.js
• Framework: Express
• RAG Provider: Exa (for source retrieval)
• LLM: TBD (OpenAI/Claude via stub interface)

Frontend (Planned)

• Will render:
  • Answer blocks with inline citations
  • Source cards with URLs and excerpts
  • Interactive evidence graph visualization

Getting Started

Backend Setup

See backend/README.md for detailed instructions.

Quick start:

cd backend
npm install
cp .env.example .env  # Add your API keys
npm run dev

The server will start on http://localhost:3001

API Example

curl -X POST http://localhost:3001/api/answer \
  -H "Content-Type: application/json" \
  -d '{"question": "What is AI transparency?"}'

Current Status

✅ Completed

• Backend scaffolding with TypeScript
• Type system for all data structures
• Stub implementations for research/answer agents
• Evidence graph builder
• Working API endpoint with mock data

🚧 Next Steps

1. Integrate Exa API for real source retrieval
2. Integrate LLM API for answer generation
3. Build frontend visualization
4. Add caching and optimization
5. Deploy to production

Development

This project was built for a hackathon with the goal of making AI more transparent and trustworthy. The focus is on:

• Clean architecture that separates concerns
• Type-safe implementation
• Clear documentation of data flow
• Extensible design for future enhancements

License

MIT