DebbieAI

Inspiration

A while ago, I learned about a product called Devin — an AI-powered software engineer that raised over $100M. However, access to Devin costs over $200/month, which inspired me to build my own service: DebbieAI — a more accessible AI software engineer (for now, mainly focused on building websites).

How I built it

The frontend is built with Next.js and TailwindCSS, designed to feel intuitive and responsive for users of all technical levels. I focused on building a clean interface where users can describe their project goals naturally.

On the backend, I used Node.js and Express.js to manage project lifecycles — from receiving prompts to deploying finished applications. When a project is submitted, it goes through a custom-built pipeline that interprets the request, constructs a project scaffold, populates it with relevant code, and packages everything into a Docker container.

A reverse proxy dynamically maps container ports to unique project URLs, making each site instantly accessible. Instead of relying solely on an AI model's output, DebbieAI applies logic to organize and verify the generated code, ensuring it's valid and deployable — like a real engineering assistant.

Challenges we ran into

One of the hardest parts was getting Gemini to not just generate code but also apply it properly — meaning creating a consistent file structure, writing to correct files, and ensuring it was deployable. Ensuring formatting consistency and output structure from Gemini was particularly tricky.