Ask about Mason.
Streaming chat backed by Claude Sonnet 4.6. The system prompt is the résumé + voice/tone guidance, cached so each follow-up is faster and cheaper. Rate-limited at 10 messages/hour per IP.
Ask me anything about Mason. Grounded on his actual career — no hallucinations about jobs he didn't have.
construction · how this works
The browser POSTs to /api/ai/chat on the FastAPI service, which streams Server-Sent Events back. The Anthropic key never reaches the browser.
The system prompt has two blocks: a tiny preamble and a curated knowledge file built from the résumé. The knowledge block is marked cache_control: ephemeral — the first request warms the cache, subsequent requests report cache_read_input_tokens > 0 and bill for ~1/10 the prefix cost.