I built a codebase-aware RAG pipeline that combines code parsing, semantic chunking, dense retrieval, BM25 hybrid search, and reranking to support repository-level code understanding.
I integrated LLM-based generation with source citations and evaluation pipelines to measure retrieval quality, answer relevance, and resistance to hallucinations. I also containerized the system with Docker and added Prometheus and Grafana observability.
For a customer intelligence and recommendation project, I built a Python and SQL analytics pipeline that turns transactional data into RFM, behavioral, and customer-level features. I developed customer segmentation, churn prediction, and personalized recommendation models, then presented customer insights in an interactive Power BI dashboard.
I also designed an MLOps workflow for experiment tracking, model versioning, reproducible training, and REST API deployment. It includes monitoring for data drift, model performance, latency, and predictions, with automated evaluation and alerting.
