Personal project · Case study
AI portfolio assistant
A conversational way for visitors to explore published experience and writing without a generic chatbot.
Problem
A portfolio should be easy to scan while still giving recruiters and collaborators a practical way to ask follow-up questions. The challenge was offering that depth without allowing an assistant to invent details.
Architecture
The site uses a small LangGraph workflow backed by Claude. Its system prompt is built from a maintained profile and the article index; tools search and read only the published Markdown posts. The chat endpoint streams responses while applying input, history, and per-IP limits.
Stack
Next.js, TypeScript, LangGraph, Claude, Markdown content, and Vercel-compatible server routes.
Implementation
The assistant is deliberately constrained to the profile and articles. When information is absent, it says so rather than guessing. The responsive sidebar stays available on desktop and becomes a compact drawer on smaller screens.
Outcome
The project is live in this portfolio as an example of a focused, grounded AI interaction. No performance or business outcome claims are made here.