Room 0901 · Floor 09 · Kaoun
Flouci Assistant
AI customer support that runs on our own model, wired into the app over WebSocket.
The problem
Support chat had to become a real second way to use the app, with AI answers, without sending customer data to a third‑party model.
What I built
- A LangChain and LangGraph pipeline on a self‑hosted model running on a rented DGX.
- Django middleware for real‑time WebSocket messaging, routing and AI orchestration.
- A PII obfuscation module that anonymises messages before the model sees them.
- The React Native chat interface: two‑way, real‑time, with AI replies inline.
- An AI layer designed as the base for future AI features in the app.