Hardik Gautam

I'm building production ML systems at IIT Roorkee, from SENTINEL, a cost-sensitive fraud detection engine, to AudioShield AI, a deepfake audio detection platform. My work combines rigorous validation, probability calibration, explainability, and end-to-end deployment.

For SENTINEL, I derived a fraud threshold from a 500:15 missed-fraud-to-false-alert cost ratio, reaching 75.7% fraud capture and 0.969 ROC-AUC on 2M transactions. I built a 61-feature FastAPI, PostgreSQL, and Docker inference pipeline with isotonic calibration and TreeSHAP explanations at under 25ms latency.

I've also trained audio-forensics models, built a MySQL star-schema e-commerce warehouse and forecasting engine, and deployed OCR, property valuation, and NLP applications. I'm seeking AI/ML Engineering and Data Science internships where I can own systems from data pipelines through deployed APIs.