On RAGAssist, my independent LLM engineering project, I built a Python/FastAPI question-answering system with document ingestion, semantic retrieval, citations, and insufficient-evidence handling. I created a labelled benchmark that reached 100% Recall@3 and improved MRR@3 from 0.815 to 0.944 through controlled comparisons.
For SupportSense, I compared NLP classifiers on BANKING77 queries and selected a baseline with 89.3% accuracy and 0.893 macro F1. In financial technology, I contributed to a military pension-board management system, debugging defects and validating expected behaviour; I also built reproducible ML training and monitoring pipelines for ModelOps.

