Back to projects
Live

AI Portfolio Assistant

A RAG chatbot that knows my whole career — retrieval, tool use, and streaming on the edge

Live on this site — try it bottom-right

AI Claude RAG Cloudflare Vectorize TypeScript Hono
Visit live View source

How it works

AI portfolio assistant (this site) The chatbot in the corner of this page. Your question gets embedded, matched against a Vectorize index of my work, handed to Claude with a set of tools it can call, and streamed straight back over SSE — all on the edge, no origin server. VisitorEdgeRetrievalGeneration question 1 · embed 2 · search 3 · context tool defs rate limit 4 · streamed tokens (SSE) You ask a question Astro Worker /api/chat KV rate limit · 3/hr Workers AI embed the query Vectorize top-k passages Tool use navigate · contact Claude composes the answer
The chatbot in the corner of this page. Your question gets embedded, matched against a Vectorize index of my work, handed to Claude with a set of tools it can call, and streamed straight back over SSE — all on the edge, no origin server.

The floating assistant in the corner of this site is a project in its own right: a retrieval-augmented chatbot that can answer almost anything about my background, projects, and skills — and route real contact messages to my inbox.

What it does

  • Answers questions about my experience, projects, and stack in my own voice
  • Retrieves the most relevant facts per question (RAG) instead of stuffing everything into the prompt
  • Uses tool calling to trigger a contact form — collect a name, email, and message, then send it
  • Streams responses token-by-token for a fast, live feel
  • Rate-limited per IP so it stays cheap and abuse-resistant

Tech stack

  • Model: Anthropic Claude (tool use + streaming) via the Claude SDK
  • Retrieval: Cloudflare Vectorize for embeddings + semantic search over a hand-curated fact base
  • Edge API: Runs on Cloudflare Workers alongside the Astro site
  • Storage / limits: Cloudflare KV for rate limiting, D1 for chat session logging
  • Abuse protection: Cloudflare Turnstile on the contact path

Why I built it

A résumé is static; a good engineer’s story isn’t. I wanted recruiters and collaborators to be able to ask — “does he know SQL Server?”, “what’s the robot actually do?”, “is he available?” — and get a specific, accurate answer in seconds. It’s also the clearest possible demo of how I build with LLMs: retrieval, tools, streaming, and guardrails, all running at the edge.

← Back to projects