An AI Chatbot on My Website
An architect should have a website. The website should have a portfolio. But what if the portfolio could talk?

The Idea
A visitor lands on the site. They want to know about my experience, stack, approach to work. They could browse pages. Or they could just ask. A live chat, except on the other end it’s not me - it’s my virtual clone, trained on my CV, projects, and publications.
Not a 2018-era chatbot with a button tree. A real conversation in natural language.
Architecture
I didn’t want to build a dumb wrapper around an API with a massive system prompt. Bio, projects, stack, publications - all of that doesn’t fit in the context window without loss. And even if it does, every request burns tokens for nothing.
Instead - RAG. All data is chunked and stored in a vector database. For every visitor question, the bot first retrieves relevant pieces via embeddings. Found three paragraphs about Kubernetes - sends only those. Asked about education - pulls only education.
The vector DB acts as a smart cache: Claude Haiku 3.5 gets a compact context instead of a full biography. Fewer tokens, faster responses, lower bills.
Stack
- FastAPI - backend, async out of the box
- Claude Haiku 3.5 - fast and cheap model for conversation
- Embeddings + vector database - relevant context retrieval
- Hetzner Cloud - hosting
- Nginx - reverse proxy
What It Can Do
The bot knows everything I know - but only the professional side. Ask about my stack - it answers. About experience in a specific area - it finds relevant projects. About my approach to architecture - it explains.
If it doesn’t know something, it says so honestly. It won’t invent projects that never existed or skills I don’t have. The system prompt strictly limits hallucinations.
Why
Three reasons:
- Conversion. Visitors get answers to their questions instantly instead of hunting through pages
- Showcase. An architect’s site that built its own AI product - better than any bullet point on a CV
- Curiosity. I simply wanted to build it and see how it works in production
Come give it a try - the chat button is in the bottom right corner at imarch.dev. Or reach out if you want one for your own site.


