At Doodhvale Farms, I optimized backend and database queries on a platform serving 10,000+ users, reducing processing time by ~40%. I also worked with user and transaction data to improve system performance and reliable data handling.
I built METO, a self-healing agentic RAG pipeline using LangGraph and Qdrant, and evaluated it with RAGAS. Across 30 queries, it achieved 0.977 Faithfulness and 0.867 Relevancy.
For GeoPulse, I built a news pipeline and trained an ensemble to assess event-driven market risk. I also applied transfer learning to CIFAR10 image classification, achieving 84.50% accuracy with a hybrid CNN-XGBoost model.
