Consilium — Multi‑Agent AI Council

Consilium is a multi‑agent AI system that simulates founder‑level strategic discussions for evaluating entrepreneurial ideas. Independent, role‑based agents analyze an idea in parallel, and a Chairperson synthesizes their insights into a single, actionable council decision.

Status: Complete and demo-ready (local execution)

Why Consilium

Most idea evaluators provide a single perspective. Consilium mirrors how real startup decisions are made — through multiple specialized viewpoints (product, market, tech, finance, risk, growth) and a final synthesis.

This project demonstrates:

• Multi‑agent system design
• Prompt orchestration and synthesis logic
• Backend API architecture using FastAPI
• Practical debugging and production‑safe file handling

---

Architecture Overview

Flow:

1. User submits an idea
2. Seven independent agents evaluate the idea (role‑specific prompts)
3. Agent outputs are aggregated internally
4. Chairperson synthesizes a final verdict
5. API returns a single structured decision

User Idea
   ↓
[7 Role‑Based Agents]
   ↓
[Chairperson Synthesis]
   ↓
Final Council Decision

---

Council Agents

Agent	Responsibility
Visionary (CEO)	Vision, mission alignment, strategic relevance
Product Lead	Problem–solution fit, MVP scope
Market Analyst	Market demand, competition, positioning
Technology Lead (CTO)	Technical feasibility and scalability
Finance Advisor (CFO)	Revenue model, costs, sustainability
Risk & Compliance	Legal, operational, execution risks
Strategy & Growth	Go‑to‑market and scaling strategy

Chairperson: Synthesizes all agent outputs into a single recommendation without introducing new analysis.

---

API Endpoints

Health Check

GET /

Returns API status.

---

Evaluate Single Agent

POST /evaluate/{agent_key}

Runs a specific agent by role.

Example agent keys:

• visionary
• product_lead
• market_analyst
• technology_lead
• finance_advisor
• risk_compliance
• strategy_growth

Request Body:

{
  "idea": "Your startup idea here"
}

---

Full Council Evaluation

POST /evaluate/council

Runs all agents sequentially and returns a synthesized council decision.

Request Body:

{
  "idea": "An AI platform that helps college students find paid micro‑internships."
}

Response (Example):

{
  "council_decision": "Council Verdict: Cautious Go..."
}

---

Tech Stack

• Backend: Python, FastAPI
• LLM: OpenAI API (model‑agnostic design)
• Configuration: python‑dotenv
• API Docs: Swagger (OpenAPI)

---

Project Structure

Consilium/
├── backend/
│   └── app.py
├── prompts/
│   ├── visionary.txt
│   ├── product_lead.txt
│   ├── market_analyst.txt
│   ├── technology_lead.txt
│   ├── finance_advisor.txt
│   ├── risk_compliance.txt
│   ├── strategy_growth.txt
│   └── chairperson.txt
├── README.md
└── PROJECT_CHARTER.md

---

Running Locally

1. Clone the repository
2. Create a .env file inside backend/:

OPENAI_API_KEY=your_api_key_here

3. Install dependencies

pip install -r backend/requirements.txt

4. Start the server

python -m uvicorn backend.app:app

5. Open Swagger

http://127.0.0.1:8000/docs

---

Key Learnings

• Multi‑agent orchestration patterns
• Safe prompt loading using absolute paths
• Debugging Python import caching issues
• Designing explainable AI systems

---

Future Enhancements (Optional)

• Parallel agent execution
• Provider toggle (OpenAI / Gemini)
• Frontend UI for idea submission
• Persistent evaluation history

---

Author

Designed and developed by Arjun R.

This project is intended to demonstrate entrepreneurship‑level system design and applied AI engineering.