Sourasish Ghosh

I've built real-time ML systems spanning fraud detection, multimodal RAG, anomaly detection, and atmospheric weather classification. My work combines Python-based machine learning with production infrastructure including FastAPI, Kafka, Redis, PostgreSQL, Docker, and vector databases.

At Spaceborn, I developed an Isolation Forest anomaly detection system for high-frequency multidimensional sensor streams and designed a low-latency Kafka, Redis, and Avro architecture for real-time inference. In my financial fraud telemetry project, I built a FastAPI, XGBoost, Redis, and PostgreSQL platform that delivered classification and SHAP reasoning in under 5ms, reaching 93.85% recall and 0.9376 PR AUC on 1.3 million rows.

I've also built an e-commerce multimodal RAG system with OpenCLIP, Qdrant, Gemini, and FastAPI, and researched hybrid fuzzy random forest weather classification using 24 years of ECMWF IFS data. At Excelerate, I led analytics work on global hiring datasets using Python, Excel, and Power BI.