CodeGen AI

A professional Python package for generating code with Google Gemini via LangChain.

Project Overview

This repository is focused on a clean code generation package and deployment-ready examples.

Included in this repository:

• codegen/ — reusable package with Gemini integration, prompt generation, validation, and auto-fix support.
• codegen_demo.py — polished command-line demo showcasing package usage.
• app_fastapi.py — REST API for remote code generation.
• app_streamlit.py — interactive Streamlit UI for generation, validation, and runtime execution.
• Dockerfile — container image definition for production deployment.
• Procfile — process configuration for platform deployment.
• .env.example — environment variable template.

Installation

Install the package in editable mode:

pip install -e .

Or install dependencies directly:

pip install -r requirements.txt

Configuration

Copy .env.example to .env and add your Gemini API key:

copy .env.example .env

Set one of the following environment variables:

• GOOGLE_API_KEY
• GEMINI_API_KEY

Running the Demo

python codegen_demo.py

CLI Usage

The package exposes a CLI entrypoint named codegen:

codegen "generate a Python function that reverses a list" --validate --auto-fix

FastAPI Server

Start the API locally:

uvicorn app_fastapi:app --host 0.0.0.0 --port 8000

Example request body:

{
  "task": "generate a Python helper that removes duplicates from a list",
  "language": "python",
  "style": "idiomatic",
  "include_docstrings": true,
  "include_type_hints": true,
  "validate": true,
  "auto_fix": false
}

Streamlit UI

Run the interactive front-end:

streamlit run app_streamlit.py

Deployment

Docker

Build and run the container:

docker build -t codegen-app .
docker run -p 8000:8000 --env-file .env codegen-app

Platform Deployment

The Procfile is included to support deployment on process-based platforms.

Package Structure

• codegen/llm_client.py — Gemini LLM abstraction with retries and structured logging.
• codegen/prompts.py — prompt builder for generation and examples.
• codegen/generator.py — code generation, validation, and auto-fix workflow.
• codegen/sandbox.py — safe Python execution sandbox.
• codegen/cli.py — CLI wrapper.
• app_fastapi.py — FastAPI application entrypoint.
• app_streamlit.py — Streamlit application entrypoint.
• codegen_demo.py — polished demo script.

Notes

• Keep secret keys out of source control and never commit .env.
• Use python -m venv .venv for isolated development.
• Update dependencies carefully when changing LangChain or Gemini SDK versions.