At 10Pearls, I built an automated system to forecast Karachi’s AQI 24, 48, and 72 hours ahead. I validated it with a 45-point real-world backtest, where 73% of forecasts were within 10 AQI points, and reported where accuracy declined rather than hiding it.
I automated hourly data ingestion and daily retraining with GitHub Actions, registered models in Hopsworks, and deployed three live model-serving endpoints. On the Pearls AQI Predictor project, I also served forecasts through FastAPI and a Next.js dashboard, addressing issues in backtesting, request performance, and live-data availability.
At Gray Mines, I built full-stack features with Next.js, React.js, NestJS, and PostgreSQL, working from UI components through REST APIs to data models. I also reproduced and traced production bugs across frontend and API layers, then verified each fix.
I built an n8n customer support agent that retrieves document content from a Pinecone knowledge base and drafts or sends Gmail replies based on confidence and urgency. I also developed React and REST API projects, including an online food ordering system and a real-time chat application.

