Chandra sai kiran Kammari

About

ML/LLM engineer with Stripe and Adobe experience who productionized a transformer-based Payments Foundation Model for real-time fraud detection at global scale (billions of transactions). Built petabyte-scale ETL/feature pipelines (Spark/EMR, Airflow, dbt, Kafka/Flink) and achieved <100ms multi-region inference (EKS, TorchServe, edge/Lambda, GPU/CPU routing) with strong PCI-DSS/GDPR compliance and explainability (SHAP/LIME), reporting a 64% fraud accuracy improvement.

Experience

Education

Key Strengths

• Built and deployed LLM/transformer-based real-time fraud detection at Stripe serving billions of transactions
• Designed petabyte-scale feature engineering and distributed ETL with Spark on EMR orchestrated by Airflow + dbt
• Achieved global <100ms inference via multi-region Kubernetes (EKS), TorchServe, edge/Lambda optimizations, and GPU/CPU hybrid routing
• Production-grade rollout practices: blue-green/canary releases with automated rollback based on latency drift and false positives
• Strong compliance and auditability focus (PCI-DSS, GDPR) using lineage/retention controls, OpenTelemetry traces, and explainability (SHAP/LIME)
• Rigorous evaluation methodology: synthetic + historical test suites, shadow-mode testing, offline/online evals, and continuous monitoring (Prometheus/Grafana/OpenTelemetry)
• Cross-functional delivery with non-technical Risk & Compliance stakeholders; translated ML concepts and defined acceptable false-positive/audit requirements
• Reported outcome: improved fraud accuracy by 64% while meeting compliance requirements