OZONE-AID

Context

OZONE-AID was built for Ozone Garage, a premium car garage, as my final-year engineering internship project at Ouerkema Innovation (iTeam University). The goal: an integrated, data-driven platform covering the full data value chain, from logging interventions to reporting back to the customer.

Problem

Fault diagnosis relied entirely on individual mechanics’ expertise, hard to standardise. There was also no consolidated view of the shop’s activity — revenue, average ticket, seasonality — to inform decisions.

Architecture

• Data lake and star-schema warehouse: collecting and modelling interventions as the shared foundation for every module.
• Assisted diagnosis (ML): symptom → likely-fault classification, cost/duration estimation (NLP, embeddings).
• RAG copilot — the project’s key differentiator: Mistral / Ollama + ChromaDB + LangGraph, to query technical documentation and assist diagnosis.
• Predictive maintenance (DL): estimating service due-dates and detecting anomalies, with automatic customer alerts.
• Decision-making dashboards (BI): KPIs — revenue, average ticket, faults by brand, seasonality, demand forecasting, no-shows.
• Customer mobile app (React Native) and back office (React / FastAPI), event-driven architecture (webhooks, Firebase/FCM push).
• Productionisation: Docker, CI, MLflow, role-based security and encryption.

Stack

Python (ML/DL), Mistral / Ollama + ChromaDB + LangGraph for RAG, FastAPI for the API, React / React Native for mobile and back office, Docker + MLflow + CI for productionisation.

Results

• Assisted diagnosis available to every mechanic, not only the most experienced ones.
• A consolidated view of shop activity through the decision-making dashboards.
• An event-driven, secured architecture (roles, encryption), ready for production.

Screenshots

Screenshots coming soon.