Personal Projects

ATLAS Pipeline — Maxibook

As a contributor to early-stage Maxibook, I built a 6-service production AI pipeline for knowledge extraction. The pipeline builds a comprehensive knowledge base starting from user intent. Includes an end-to-end orchestrator with step caching and retry logic. Built with FastAPI, LangChain, Pydantic, Tavily search, and LLM-as-a-judge evaluation. Deployed on AWS EC2 via systemd.

Distributed Object Detection — CS5939

Cloud-Edge system architecture for real-time object detection across two university VMs. Flask API serving YOLO on a cloud node; edge client with a live Dash dashboard on a second node. Includes an automated experiment runner collecting performance results across configurations.

End-to-end ML pipeline

EDA, ML model selection, experiments and deployment. Deployment is based on command-line serving of pre-trained Random Forest models via argparse and joblib. Accepts JSON input, routes to the correct stacked model based on cycle count, and returns predictions as JSON. A clean example of wrapping a pickled model in a deployable CLI interface.

London Fire Brigade — Geospatial Analysis

Loaded and pre-processed real London Fire Brigade incident data using PySpark, then built a Folium choropleth map visualising fire incident density across London boroughs. Combines distributed data processing with interactive geospatial visualisation.