Introduction
Code Generator is an AI agentic prototype of a software engineer. Given a public GitHub repository and specific prompts, it can automatically fix issues, build new features, and test code. I originally built this project during a startup engineering competition (I was a finalist), but then expanded it further on my own time.
The service outputs the difference (diff) between the original and transformed code. Key technologies used include FastAPI and Python to build the API service. I fine-tuned ChatGPT-3.5 (the latest model at the time) on example prompts, code, and solutions, while also utilizing custom chained prompts and external context to enhance output accuracy.
I hosted the service on an AWS EC2 instance, using NGINX as a reverse proxy for live access on the internet. Below, I’ve included additional resources such as documentation, a Postman guide, a system design diagram, and other relevant materials.
Due to ChatGPT’s new pricing model (no free tier), the service is currently down.
Run in Postman:
Key Documents & Links
🗄️ GitHub
📄 README
How To Use TLDR