ask Hiten
A small live demo: ask anything about my work, stack, side projects or availability. Answers come from free hosted open-weight models (GPT-OSS 120B primary, Llama 3.3 70B failover), grounded in a short CV I wrote by hand. Honest by design: it will say so when it doesn't know.
Built with
Everything powering this page, end to end. No framework runtime in the browser, just a tiny streaming client and a Node process behind a TLS edge.
- Astro 5 (SSR) server output + @astrojs/node standalone
- TypeScript strict-mode, no client framework
- Vanilla streaming client fetch + ReadableStream, parses OpenAI SSE
- Node 22 on Alpine container on the Oracle ARM VM
- GPT-OSS 120B Groq/Cerebras primary; NVIDIA / OpenRouter / Together (Llama 3.3 70B) failover
- Provider failover chain health-checked at SSR, fails through automatically
- In-memory rate limit 20 req / 10 min per IP, token-bucket
- Forgejo Actions CI self-hosted runner: build → rsync → restart
- WireGuard tunnel NAS source → Oracle VM deploy target
- Traefik + Cloudflare DNS-01 TLS without opening home ports