AI Software Company Simulator

A multi-agent AI system that simulates an entire software company. Give it any software requirement — 5 specialized AI agents collaborate to plan, design, build, and review it.

Built from scratch using sequential multi-agent orchestration — no frameworks.

---

How it works

Your requirement
      ↓
  CEO Agent — breaks requirement into tasks and priorities
      ↓
  PM Agent — writes user stories and feature spec
      ↓
  Designer Agent — creates UI/UX design and user flow
      ↓
  Developer Agent — writes actual working code
      ↓
  QA Agent — reviews code, finds bugs, writes test cases
      ↓
  Full report downloaded as .md

Each agent has its own prompt, its own role, and its own output — they pass context to each other in sequence. This is multi-agent orchestration.

---

Demo

Type any requirement → watch 5 agents work in sequence → get a full project output with code, design, and test cases.

---

Project structure

ai-software-company-simulator/
├── app.py              # Streamlit UI
├── ceo_agent.py        # CEO — project plan + task breakdown
├── pm_agent.py         # PM — user stories + feature spec
├── designer_agent.py   # Designer — UI/UX + wireframe
├── dev_agent.py        # Developer — actual working code
├── qa_agent.py         # QA — code review + test cases
├── prompts.py          # All agent prompts in one place
├── config.py           # API keys from .env
├── save_output.py      # Saves full report as .md
├── .env.example
└── requirements.txt

---

Tech stack

Layer	Tool
LLM	GPT-4o-mini (OpenAI)
Frontend	Streamlit
Agent pattern	Sequential multi-agent (built from scratch)
Output	Markdown report
Config	python-dotenv

---

Setup

1. Clone the repo

git clone https://github.com/ES7/ai-software-company-simulator
cd ai-software-company-simulator

2. Install dependencies

pip install -r requirements.txt

3. Set up API key

cp .env.example .env

Fill in .env:

OPENAI_API_KEY=your_openai_key_here

4. Run

streamlit run app.py

---

Example requirements to try

• "Build a REST API for a todo app with user authentication and PostgreSQL"
• "Create a Telegram bot that summarizes YouTube videos"
• "Build a dashboard that tracks crypto prices in real time"
• "Design a recommendation system for an e-commerce platform"

---

What each agent produces

Agent	Output
CEO	Project name, task breakdown, priorities, success criteria
PM	User stories, feature spec, tech recommendations, risks
Designer	User flow, UI components, color scheme, wireframe description
Developer	Working code files, setup instructions, architecture notes
QA	Bug report, test cases, security concerns, approval status

---

What I learned building this

• How to orchestrate multiple agents sequentially — each agent's output becomes the next agent's input
• How different system prompts create genuinely different agent personalities and outputs
• How to handle JSON parsing failures gracefully when LLMs generate code with special characters
• That multi-agent systems produce better output than a single agent doing everything

---

Author

Ebad Sayed — Final year, IIT (ISM) Dhanbad, Co-founder of Voke AI

Connect: LinkedIn · GitHub · X