This Site

“I build with AI” is easy to claim and hard to verify on a portfolio. I wanted visitors to try the work, not read a screenshot of it. The catch is that a public, in-browser AI demo creates two real risks: leaking the API key to anyone who opens dev tools, and letting strangers spend your quota. I treated those as the actual design problem.

A streaming Gemini chat that renders the response token by token, so the interface feels live instead of frozen while it waits.

A text analyzer that returns typed, structured JSON using structured-output mode, so the result is data the UI can render reliably rather than a string to guess at.

Serverless route handlers that keep the API key server-side, with rate limiting on both demos.

Server-side route handlers instead of calling the model from the browser. Calling from the browser is faster to build and a common shortcut. It also exposes the key and the quota to anyone. Routing through serverless functions cost a little more plumbing and bought the only acceptable security posture for a public demo.

Structured-output mode over free-text parsing. Asking for typed JSON constrains what the demo can show. It also means the UI never breaks on a malformed response, which is the difference between a demo that works in front of a stranger and one that does not.