Ask Bin Anything
A RAG agent over my public writing — ask it things when I'm not around.
This is an AI agent trained on Bin's public writing. It may be wrong. For anything important, email him directly.
How it works
- 01
Your question becomes a vector
The text is embedded, then matched against chunks of my writing — experience timeline, blog posts, about page.
- 02
Top snippets pack into a prompt
The four most relevant chunks are assembled with instructions to answer in your language and cite sources inline as [n].
- 03
A concise answer streams back
A non-reasoning language model replies in 2–3 sentences with inline citations you can click to verify.
Under the hood
Next.js 15 with content indexed into SQLite as embeddings at build time. Per-IP rate limit (10/hour), site-wide cap (100/day), 24-hour answer cache. An automatic provider-fallback chain keeps the widget answering even when the primary model is rate-limited or down.
Honest limits
It only knows what I've published on this site. For anything private, recent, or important — email me directly. It sometimes misremembers dates or paper titles, so click the cited sources if you need to be sure.