Tools
Ask my portfolio
A tiny AI that runs on your own graphics card. Nothing is sent to a server.
Ask about my projects, skills, education or blog posts. The answer comes only from a small public context file of this site (context.md), found by keyword search. Anything outside that gets a polite refusal.
1. Search works right now
Please note: the AI is a very small model trained on limited data and it can make mistakes. It only reads the short context file of this site, so please verify anything important with the linked pages or by contacting me.
Without the AI model, you still get the best matching passages from the index. No download, no GPU needed.
2. Optional: let the AI write the answer
Big download warning. Turning this on downloads a small language model (Llama 3.2 1B, about 0.7 to 0.9 GB) into your browser cache, then runs it with WebGPU. Use Wi-Fi. It is saved after the first time, and you can clear it from your browser site data.
AI mode: off. Search mode only.
Data sources & privacy
Your questions never leave your browser. Search runs on index.json, a file built from this site's public pages. If you opt in, the page loads the WebLLM library from cdn.jsdelivr.net (pinned to version 0.2.85 and checked against a SHA-384 hash before it runs) the model's compiled code from raw.githubusercontent.com and the model weights from huggingface.co; each of these services sees your IP address like any download. Those two model sources are not pinned to a fixed version, which is one reason this stays an opt-in experiment. Small models make mistakes, so check anything important against the linked source.