AI Startup Builder Agents

A small FastAPI service that takes a one-line startup idea and returns a draft startup plan: a market and competitor analysis, a brand name and tagline, a business model, and a set of financial calculations with short recommendations. Each part is produced by a separate Python "agent" module; the text-generating agents call the Google Gemini API (gemini-1.5-pro).

How it works

api/main.py exposes one main endpoint, POST /create_startup, which calls the agents in sequence and returns their combined output as JSON:

1. Idea processing (agents/input_agent.py): get_startup_idea echoes the idea with a status message. It is a placeholder and makes no model call.
2. Market research (agents/market_research_agent.py): analyze_market_trends asks Gemini for the top three competitors with similarity scores, then splits the response into name / similarity_score pairs.
3. Brand identity (agents/brand_identity_agent.py): generate_brand_name_and_tagline asks Gemini for a brand name and tagline and returns the raw text.
4. Business model (agents/business_model_agent.py): create_business_model asks Gemini for revenue streams, cost structure and target customers, and parses the response into three lists.
5. Financial analysis (agents/financial_agents.py): five functions (optimize_product_sales, optimize_subscription_service, calculate_development_cost, calculate_marketing_cost, calculate_operations_cost) compute simple figures such as projected revenue, churn loss, cost per feature and remaining marketing budget in plain Python, then ask Gemini for a short recommendation on each.

agents/financial_insights_agent.py contains derive_business_insights_from_financial_data, which aggregates the financial outputs into profit margin, breakeven status, marketing efficiency and a scalability factor, and adds rule-based recommendations without calling a model.

GET / is a health check.

Repository layout

startup-builder-agents/
├── api/
│   └── main.py                      FastAPI app and the /create_startup pipeline
├── agents/
│   ├── input_agent.py               idea intake (placeholder)
│   ├── market_research_agent.py     competitor analysis via Gemini
│   ├── brand_identity_agent.py      brand name and tagline via Gemini
│   ├── business_model_agent.py      revenue, costs, customers via Gemini
│   ├── financial_agents.py          revenue and cost calculations with Gemini recommendations
│   └── financial_insights_agent.py  rule-based summary of the financial outputs
└── models/                          empty package

Getting started

Prerequisites

• Python 3
• A Google Gemini API key

Install

The repository does not include a requirements file. The code imports the following packages, and needs an ASGI server such as uvicorn to run:

pip install fastapi uvicorn pydantic google-generativeai python-dotenv

Configure

The agents call load_dotenv() and read one environment variable:

Variable	Used for
GEMINI_API_KEY	Authenticating calls to the Gemini API

Export it in your shell, or put it in a .env file in the directory you start the server from.

Run

Start the server from the repository root, so that the agents package is importable:

uvicorn api.main:app --reload

Usage

curl -X POST http://localhost:8000/create_startup \
  -H "Content-Type: application/json" \
  -d '{"idea": "A subscription service for refurbished laptops for students"}'

The response is a JSON object with these top-level keys:

• idea_processing
• market_analysis (market_size, growth_potential, competitor_analysis)
• brand_identity
• business_model (revenue_streams, cost_structure, target_customers)
• financial_analysis (product_sales, subscription_service, development_cost, marketing_cost, operations_cost)

If any step raises, the endpoint returns HTTP 500 with the error message.

Status / limitations

• The financial inputs (price, units, subscribers, churn rate, budgets and so on) are hard-coded in api/main.py; they are not derived from the submitted idea or taken from the request.
• market_size and growth_potential in the market analysis are fixed placeholder strings; only the competitor list comes from the model.
• Model responses are parsed with simple line splitting, so the structure of competitor_analysis and the business model lists depends on how the model formats its answer.
• derive_business_insights_from_financial_data is implemented but not called by the API.
• There is no persistence, authentication, test suite, or container configuration in the repository.