Live project
AI Portfolio Assistant
An AI chat assistant built into this portfolio that answers visitors' questions about my projects, experience, and availability. It runs on Google's Gemini API (free tier), streams answers word by word, and only answers from my resume and project data, so it doesn't make things up.
- Technologies
- 5
- Key features
- 6
- Deployed
- Live
The Problem
Recruiters and potential clients often have one quick question, like what stack I use or whether I'm available, and won't read every page to find it.
My Role
Sole developer: designed the chat UI, wrote the Next.js route handler that calls the Gemini API, built the knowledge base from the site's own project data, and added the safeguards for a public endpoint.
Key Features
- Streaming answers rendered as they are generated
- Knowledge base generated from the same project data the site uses, so it stays in sync
- Instructions to answer only from that data and say so when something isn't covered
- Low thinking level and a small output limit for fast, short answers
- Input limits, short conversation history, and per-visitor rate limiting
- Suggested questions and a graceful fallback to WhatsApp if the assistant is unavailable
Challenges
A public AI endpoint can be spammed and the free tier has daily limits, so the main work was keeping it bounded: capped message length and history, a small output limit, low thinking level for short factual answers, per-visitor rate limiting, and a friendly fallback to WhatsApp when the limit is hit.
