Replying to WhatsApp all day

Replying to WhatsApp all day is draining, but letting an AI answer everything is risky. I wanted a system where routine messages are handled automatically and sensitive ones wait for human approval.

A control dashboard connected to a zero-cost n8n backend. Incoming text, voice notes and images are transcribed or described by AI. The assistant either replies automatically or holds the message for review.

How it works:

Voice notes are transcribed with Groq Whisper, and images are described with Gemini 2.0 Flash.

A Gemini-powered agent decides between “auto-send” and “held for review”.

Replies can go out as native WhatsApp voice notes using Google Text-to-Speech.

The dashboard shows the live message stream, an approval queue, switches for Auto-Pilot, Voice Synthesis and Intervention Mode, and an animated voice visualizer.

Stack: React, Tailwind CSS, Framer Motion, n8n, WhatsApp Cloud API, Groq, Gemini, Google TTS.

What I learned: Designing the approval flow mattered more than the AI itself. Deciding what the assistant is not allowed to send on its own shaped the whole design.