Building reliable AI systems from experimentation to production impact.

• Developing graph-based trade surveillance for ~3.2M daily equity trades and improving recall on flagged cases by 34%.
• Built Kafka/Flink anomaly detection modules on ~200K events/sec streams with sub-second detection latency.
• Implemented explainability workflows (SHAP + attention methods) aligned with SR 11-7 model-risk compliance.

• Led 5 engineers to build LLM-powered patient-trial matching and ad optimization systems.
• Improved matching precision by 50% and doubled qualified conversions while reducing spend by 30%.
• Implemented production monitoring for CTR, CPL, drift, and fairness in a HIPAA-aligned setup.

• Built multi-label BERT classifier on 2.3M tickets with 91.4% micro-F1 and 68% MTTA reduction.
• Deployed anomaly and escalation prediction services using Kafka + Isolation Forest + XGBoost.
• Implemented Kubeflow/MLflow and data quality checks, reducing production failures by 43%.

Built an end-to-end RAG system for 10K+ documents with semantic retrieval and low-latency inference; reduced manual investigation effort by ~60%.

Developed a GPU-accelerated tracking system with 0.97 F1 and reduced latency by ~30%, sustaining 20–25 FPS.

Implemented uplift modeling (T/S learners + A/B testing), achieving +18% conversion lift and reducing low-value outreach by 22%.