Your WhatsApp pen pal is a vector database

What it does

Ava is a six-lesson course that walks you through building a multi-modal WhatsApp agent. The resulting bot can handle text, voice messages, and images over WhatsApp, and it maintains long-term memory in Qdrant so it can recall details from past conversations. It also generates images of its own “daily activities” to send back to you, because even agents need hobbies.

The interesting bit

Instead of yet another chatbot demo, this is framed as a production-oriented integration project: you wire together Groq for fast inference, ElevenLabs for voice, Together AI for image generation, and LangGraph to orchestrate the workflow. The agent invents a persona with ongoing life updates, turning a standard API integration into something closer to a Turing Test prank for your contacts.

Key highlights

• Multi-modal over WhatsApp: handles text, speech-to-text via Whisper, vision via Llama 3.2 Vision, and text-to-speech via ElevenLabs
• Long-term memory using Qdrant vector database to recall past conversation details
• Image generation pipeline using FLUX models through Together AI to create activity updates
• LangGraph workflows for state persistence and agent orchestration
• Deployable to Google Cloud Run with WhatsApp API integration
• Designed to run on free tiers from Groq, ElevenLabs, Qdrant Cloud, and Together AI

Verdict

Good for software and ML engineers who want a guided, end-to-end integration project rather than a black-box framework. Skip it if you are looking for a ready-to-deploy WhatsApp bot without touching LangGraph or managing half a dozen API keys.