Things I've
built.

Spent about 48 hours building an AI design tool that turns sketches into actual visuals. I've always had this problem: you can picture something in your head, maybe even scribble it out, but turning it into a real design is a challenge. That's where things fall apart. So for the NanoBanana Hackathon, I wanted to build something that addresses exactly that. Inspacio takes your rough canvas drawings, runs them through Google's Gemini API, and generates polished visuals from what you sketched. Add text instructions if you want more control, or just let the drawing speak for itself. Built the whole thing with vanilla JS with libraries, HTML, and CSS because speed mattered more than fancy frameworks. Canvas input is converted to base64 PNG and processed automatically. The hardest part was, honestly, the prompt engineering to make the AI understand terrible drawings. The surprising part? It actually works. From OCR -> Image/Text Parsing -> Encoding into base64 string for API Parameter -> Gemini API - Output Image What this really showed me is that the gap between having an idea and seeing it come to life is shrinking fast. You don't need design chops anymore, just the ability to communicate what you want. (Website works best on Laptop/PC/Mac)

1. Engineered and operated a high-concurrency game scaling to 19M+ lifetime visits, $11,000+ USD in revenue, and 1.6K peak concurrent users with 99.9% uptime. 2. Built datastore-backed progress persistence (change-detection on save to cut redundant writes), a settings/menu system, and global leaderboards with country-specific speedrun modes to drive competitive engagement. 3. Designed a checkpoint/teleport system compatible with StreamingEnabled: since clients can’t see distant unloaded checkpoints, routed teleports through remote functions to resolve them server-side — trading minor network cost on rare far-jumps for the performance gains of streaming.