Nithin Gowda P

I’m an AI/ML Engineer focused on building production-grade AI systems across machine learning, deep learning, and Generative AI—backed by scalable backend infrastructure. My work emphasizes reliable inference, evidence-grounded answers, and explainability that teams can trust.

In my AI/ML Intern role at IIMSTC, I engineered an electricity consumption prediction pipeline using 5-fold cross-validation, reaching R of 0.90 and a 29% improvement over baseline. I also processed 10,000+ records, where XGBoost reduced prediction variance and improved consistency across high-consumption and spike/edge cases.

On my project side, I built a CKD Clinical AI Decision Support System that combines Logistic Regression (AUC 0.98), SHAP explainability, and RAG-driven evidence retrieval, wrapped in HITL reliability workflows. I optimized async FAISS retrieval to 166ms within a 2.6s end-to-end pipeline and added JSON audit trails for traceability and governance.

I’ve also delivered low-latency, high-concurrency ML services—like a Transaction Risk Scoring API using async FastAPI with sub-150ms latency and 160 RPS (Locust) stability. Beyond projects, I contributed to Haystack (deepset) by designing a retrieval confidence scoring enhancement to better distinguish retrieval failures from LLM-generated errors in multi-query RAG workflows.