DEV Community

Hung Tran
Hung Tran

Posted on

Building an Open Source Project with 100% AI-Generated Code

Introduction

I wanted to customize an LLM by wrapping it with a system prompt that I define to create an API for specific processing. This led me to build this open-source project.

About This Open Source Project

This project allows you to create a backend API wrapping an LLM in just one minute. All you need to do is enter the API path, request schema, response schema, and system promptβ€”then it's ready to go!

How I Built It

Technology Stack

  • Frontend: React + Vite, because I specialize in React, and Vite offers fast builds.
  • Backend: Node.js.
  • Database: PostgreSQL, as I am familiar with it.
  • Migrations: Knex.js (I had never used it beforeβ€”AI generated it for me).

Tools Used

  • Cursor IDE combined with Claude 3.7 Sonnet.
  • Custom Rules for precise data handling, sourced from: Awesome Cursor Rules.

Claude 3.7 sonnet is the best

Prompting Strategy

I used role-based prompting to make AI act as both a Senior Developer and a UI/UX Designer, ensuring a modern and aesthetically pleasing UI.

Role-based prompting

Development Steps

  1. Define AI agent roles: I described the AI agent's role and the project requirements, then asked it to initialize the frontend and backend source code.
  2. Frontend development: Since the role was predefined, I directly described the required features and workflows to generate the frontend.
  3. Backend development: I instructed the AI to read the entire frontend codebase and generate a corresponding backend. Surprisingly, it generated exactly what I expected!
  4. Docker deployment: I asked AI to generate Docker configurations.
  5. Frontend-Backend Integration: I provided a prompt to make AI integrate the generated backend API into the frontend.
  6. Testing & Debugging: Finally, I manually tested the project and asked AI to fix the bugs. This step could be automated, but I chose to review manually.

Results

  • 100% of the code was generated by AI.
  • I made approximately 150 requests to Claude 3.7 sonnet to complete the project.
  • The entire development process took 3 hours, with an additional 2 hours for bug fixes (still fully AI-driven).

English grammar checker API demo:

Javascript mentor API demo

Conclusion

Building a fully AI-generated open-source project is not only fast but also entirely feasible. However, this poses a serious risk to junior developers, as AI can potentially replace them. Two years ago, no one could have imagined AI reaching this level. I strongly believe that in a year, AI will be capable of replacing mid-level developers. In the near future, a large project may only require one senior developer and multiple AI assistants to be successfully completed.

Open Sourcing and Community Support

I have open-sourced this project on GitHub: https://github.com/hungxplorer/ez-ai-agent

If you find it useful, please give it a ⭐ so I can continue building more AI-powered projects. Thank you!

Top comments (0)