PAI (Portfolio AI) is the primary interface to this studio site — not a chatbot bolted onto pages, but the chat. Visitors ask in natural language; the system retrieves grounded knowledge, opens work in a sidebar panel, and responds with depth matched to intent.
A traditional portfolio shows the same content to every visitor. PAI routes — recruiter vs. practitioner vs. founder, overview vs. process depth, voice AI to Andesite, gated work to an access request without qualifying questions.
Outcomes.
Conversation as navigation
Questions replace browse literacy — the visitor doesn’t need to know where the work section lives.
Show, don’t tell
Six panel types surface real portfolio pages, kits, and generated HTML mid-thread — not prose descriptions of links.
Keyword corpus at scale
~3,945 chunks, intent modes, always-include pinning, and live annotations — retrieval design, not embeddings by default.
Voice that reads as speech
STT proxy, multi-provider TTS, and a normalization pipeline tuned for pace and pause — not text-to-speech flatness.
What the build notes cover
- Streaming architecture, dual LLM providers, and SSE panel events
- Knowledge design, ingest modes, and prompt rule blocks
- Page proxy, panel state loopback, and trust boundaries for gated work
- Personalization — greetings as code, geo hints, per-company
?ctx=URLs