Ameya Bhalgat

Built scalable, fault-tolerant data pipelines in Python orchestrated with Apache Airflow to automate ETL workflows for multi-modal fusion data (sensor time-series, imaging, magnetic fields), enabling feature extraction and improving processing efficiency by 40%. Trained and validated real-time anomaly detection and forecasting models (LSTM, Autoencoders, Isolation Forest) on noisy time-series data with calibrated thresholds, improving breakdown detection and experimental safety by 30%. Engineered low-latency feature extraction and event-triggering logic using signal-processing and time-series analysis techniques (cross-correlation, peak detection, convolutional filtering) to support real-time ML inference in safety-critical systems.

Developed AI-driven predictive fraud detection model using Random Forest and GBM, along with NLP techniques (TF-IDF, Word2Vec, BERT) to analyze transaction data, reducing fraud by 35% and improving security for 10,000+ users. Built a trust-based recommendation system using collaborative filtering and behavioral features, improving trust scoring accuracy by 40%, with model experimentation and dataset management supported via AWS (EC2, S3).

Optimized a Nifty Futures trading strategy by enhancing an XGBoost model using LLM-assisted feature ideation workflow and automated back-testing, improving prediction accuracy by 5.3% and increasing annualized returns by 18%. Designed an agentic experimentation workflow where an LLM supported feature hypothesis generation and experiment comparison across market regimes, enabling rapid testing of 57 engineered features, reducing pipeline processing time by 8%.