A-Gen 🚀

Multi-Agent Software Engineering Assistant

A-Gen is an Agentic AI-powered software engineering assistant that transforms natural language requirements into complete software projects through a collaborative multi-agent workflow.

Instead of relying on a single LLM call, A-Gen mimics a real software development team by orchestrating specialized AI agents responsible for planning, architecture design, and implementation.

---

Overview

A-Gen takes a prompt such as:

Create a modern weather application using HTML, CSS, and JavaScript.

and automatically:

1. Analyzes the requirement
2. Creates a structured project plan
3. Designs the project architecture
4. Breaks work into implementation tasks
5. Generates code file-by-file
6. Creates a complete project structure

A-Gen Multi-Agent Workflow: Planner → Architect → Coder → Generated Project

---

Key Features

• Agentic AI Workflow
• Multi-Agent Architecture
• Autonomous Task Decomposition
• Structured Output Generation
• Tool Calling
• Stateful Workflow Orchestration
• Multi-File Project Generation
• React Application Generation
• HTML/CSS/JavaScript Project Generation
• LangGraph State Management
• Groq LLM Integration
• Automated File Creation

---

Architecture

User Prompt
     │
     ▼
┌─────────────┐
│ Planner     │
└─────────────┘
     │
     ▼
┌─────────────┐
│ Architect   │
└─────────────┘
     │
     ▼
┌─────────────┐
│ Coder       │
└─────────────┘
     │
     ▼
Generated Project

---

Agent Workflow

Planner Agent

Responsibilities:

• Understand user requirements
• Identify project scope
• Define features
• Determine technology stack
• Generate file structure

Output:

{
  "name": "Weather App",
  "techstack": "HTML CSS JavaScript",
  "features": [
    "Weather Search",
    "Forecast",
    "Responsive Design"
  ]
}

---

Architect Agent

Responsibilities:

• Analyze project plan
• Break project into implementation tasks
• Create dependency-aware execution order
• Define file responsibilities
• Create implementation roadmap

Example:

Task 1:
Create index.html

Task 2:
Create style.css

Task 3:
Create script.js

Task 4:
Integrate API functionality

---

Coder Agent

Responsibilities:

• Read existing files
• Generate production-ready code
• Maintain consistency across files
• Implement project features
• Write code directly to disk

Uses:

• write_file()
• read_file()
• list_files()
• get_current_directory()

---

Tech Stack

AI & Agent Frameworks

• LangGraph
• LangChain
• Groq
• Pydantic

Programming Language

• Python

Workflow Components

• Structured Outputs
• Tool Calling
• State Management
• Multi-Agent Orchestration

---

Project Structure

A-Gen/
│
├── agents/
│   ├── graph.py
│   ├── prompt.py
│   ├── states.py
│   ├── tools.py
│   └── __init__.py
│
├── generated_projects/
│
├── .env
├── main.py
├── pyproject.toml
└── README.md

---

State Management

A-Gen uses LangGraph state management to track:

• User Prompt
• Project Plan
• Task Plan
• Current Step
• Generated Files

This enables agents to collaborate across multiple workflow stages.

---

Tool System

write_file()

Creates or updates files inside generated projects.

read_file()

Reads existing file content.

list_files()

Lists files generated by the workflow.

get_current_directory()

Returns active project directory.

---

Generated Project Storage

Each generated application is stored inside:

generated_projects/

Example:

generated_projects/
│
├── weather-app/
├── calculator-app/
├── portfolio-website/
└── expense-tracker/

---

Installation

Clone Repository

git clone <repository-url>
cd A-Gen

Create Virtual Environment

uv venv

Activate Environment

Windows:

.venv\Scripts\activate

Linux/macOS:

source .venv/bin/activate

Install Dependencies

uv sync

---

Environment Variables

Create a .env file:

GROQ_API_KEY=your_api_key_here

---

Running A-Gen

uv run python -m agents.graph

or

python main.py

---

Example Prompts

Create a modern weather app using HTML, CSS and JavaScript.

Create a responsive portfolio website.

Create a React-based expense tracker.

Create a task management dashboard.

Create a Netflix landing page clone.

---

Future Improvements

• Reviewer Agent
• Debug Agent
• Self-Healing Workflows
• FastAPI Backend
• Next.js Frontend
• ZIP Export
• Deployment Automation
• Multi-LLM Support
• Memory Layer
• Human-in-the-Loop Review

---

Resume Highlights

A-Gen demonstrates:

• Agentic AI
• Multi-Agent Systems
• LangGraph
• LangChain
• Workflow Orchestration
• Tool Calling
• State Management
• Structured Outputs
• Autonomous Software Development

---

Author

Ananya Gupta

Built as an exploration of Agentic AI, Multi-Agent Systems, and Autonomous Software Engineering.