import caso from "cv/proyectos/asistente-whatsapp-ia";
AI WhatsApp assistant that knows when to hand off
caso.summary;
A WhatsApp assistant that answers with an LLM and an editable knowledge base, transcribes voice notes on the server, understands images and hands off to a person when needed. In production at the congress from September to October 2026; second deployment in November 2026 for a direct-sales brand, in listen-only mode.
caso.role;
My part
Sole designer, developer, tester and operator, including deployment.
caso.highlights.map(h => <li>{h}</li>);
What it includes
- Voice notes transcribed with whisper.cpp on the server: audio never leaves the machine.
- Read-only registration lookups scoped to the sender's number; flight details are saved only after an explicit "yes".
- Handoff to staff with a summary, and automatic pause when a human steps in.
- Daily learning loop applied only after human approval.
- A 37-case suite that checks whether the bot answers or hands off, plus message-filtering tests.
- Second deployment: interest classification (interested / hot), a single follow-up and team alerts with the source ad.
caso.stack;
Stack
- TypeScript
- Bun
- LLM
- whisper.cpp
- ffmpeg
- SQLite
- Docker
- Dokploy
export default <Acciones />