About

• Search relevancy in AI-aided information retrieval

• Built Unreal Data, a domain-agnostic engine for generating synthetic 2D data using reward-maximization (Reinforcement Learning) — recipient of IBM's Hyper Blue program grant
• Owned the AI/ML engine (DQN algorithm, TF2) and MDP environment design for tabular data, extending the OpenAI Gym environment
• Ran performance testing for the Ray distributed framework on AWS

• Predicted insurance fraud (Auto, Property Casualty, Workers' Compensation) using tabular and unstructured data
• Owned feature engineering, risk modelling, hypothesis testing, and model explainability (SHAP)
• Productionized pipelines and models using PySpark, spark-nlp, and IBM CP4D

• Designed and developed an entity resolution framework for detecting financial crimes using probabilistic matching
• Built the foundation for AML (Anti-Money Laundering) and Surveillance Insights analytics
• Performance-tuned watchlist processing on the HDP platform

• Worked on image classification, segmentation, and object detection using ResNet, YOLO, and GANs

• Architected video analytics for perimeter intrusion detection at a major steel plant
• Architected a sensor-driven flood forecast and early warning system for a South-East Asian government
• Led technical consulting across Australia, EU, Middle East, South-East Asia, and the US

• Design and development lead — merge ownership, defect tracking, and release ownership

A domain-agnostic engine for generating synthetic 2D data using reward-maximization (Reinforcement Learning) to mimic realistic decision-making. Recipient of IBM's Hyper Blue program grant.

Probabilistic matching system for finding obvious and non-obvious relationships between records, powering AML and Surveillance Insights analytics at IBM.

Sensor-driven flood prediction and early warning system architected for a South-East Asian government, integrating water level, rainfall, video, and GIS/RADAR data.