helloio
ActiveAI integration · Claude + RAG on pgvector
AI scoping assistant: architecture and build
We replaced the intake form on the helloio site with a conversation. Visitors describe a rough idea; the assistant guides them, one key question at a time, then compiles an actionable project brief.
Answers are grounded in retrieved knowledge of real cases and services — no invented estimates, no inflated claims. The system runs on the same stack we use when bringing AI to clients: Claude, pgvector, Next.js.
Getting an AI to talk is easy. The hard part is everything behind it: retrieval feeding it facts, guardrails containing cost, an intake pipeline catching the result. The model is one layer; the system is the work.