Building the First Version of SiteGen AI – Turning an Idea into Reality

A developer built the first version of SiteGen AI, an application that generates websites from user prompts using FastAPI and the Gemini API. The initial prototype, though imperfect, proved the concept and led to lessons about the importance of prompt engineering and system design. The project has since evolved into a vision for an AI Software Engineer capable of generating complete applications.

In my previous article, I shared how the idea for SiteGen AI was born. In this article, I want to talk about how I built the first working version of the project. Spoiler alert... It wasn't perfect. But it worked. And that's where every great project starts. The first version had only one objective. A user types something like: Build a modern portfolio website for an MCA student. The application should automatically generate: and display the result instantly. Sounds simple. It wasn't. Since I'm comfortable with Python, I decided to use FastAPI for the backend. My initial stack looked like this. I intentionally kept the first version simple. The goal wasn't to build the perfect architecture. The goal was to make something work. The first workflow looked like this. User Prompt ↓ FastAPI Backend ↓ Gemini API ↓ Generated HTML ↓ Browser Preview The user entered a prompt. The backend sent it to the AI. The AI generated the website. The browser rendered it immediately. When I saw my application generate its first webpage, it felt amazing. One of the first prompts I tested was: Build a portfolio website for a software developer. Within seconds, the application generated: HTML CSS JavaScript It wasn't beautiful. It wasn't production ready. But it proved that the idea worked. As I continued testing, I noticed several issues. Sometimes the generated websites looked almost identical. Sometimes the AI ignored parts of the prompt. Sometimes the design was too generic. The more prompts I tested... the more problems I discovered. At first I thought the AI was the problem. Later I realized the real problem was my architecture. One of my biggest lessons was this: Good AI output depends on good prompts and good system design. Simply sending: Build a business website isn't enough. The AI needs context. It needs structure. It needs clear instructions. That realization completely changed how I thought about AI development. The first version of SiteGen AI was only the beginning. It could generate simple websites. But my vision became much bigger. I didn't want to build another HTML generator. I wanted to build an AI Software Engineer . A system capable of generating complete applications, not just webpages. That vision became SiteGen AI v2 . I'll talk more about that in the next article. Looking back, the first version wasn't impressive from a technical standpoint. But it taught me one of the most important lessons in software engineering: Start simple. Learn. Improve. Repeat. Every feature I add today is built on top of that first working prototype. If you're building your own AI project, don't wait until it's perfect. Build the first version. You'll learn far more from a working prototype than from endless planning. Thanks for reading If you're also building AI applications, I'd love to hear about your journey in the comments. Happy Coding