Sumit Sahu

About

ML/AI engineer who deployed a real-time, edge-based computer-vision pipeline for produce recognition in retail self-checkout to reduce shrink. Demonstrates strong end-to-end production chops: multi-camera data calibration/sync, ranking-based modeling for fine-grained classes, latency-focused optimization, and continuous A/B testing/monitoring with guardrails. Experienced with ML orchestration (Kubeflow Pipelines, Airflow) and CI/CD via GitHub Actions, and collaborates closely with store operations to make interventions usable in the checkout flow.

Experience

Education

Key Strengths

• Built and deployed production computer-vision system for retail self-checkout produce recognition to reduce shrink
• Deep experience handling noisy multi-camera data (preprocessing, calibration, synchronization) in real-world environments
• Improved model performance on visually similar classes using ranking-based approach and a separate CPG-vs-produce model
• Edge deployment under tight latency budgets with model/batching/integration optimizations for real-time inference
• Strong evaluation and reliability practices: unit + synthetic tests, benchmarks, shadow/A-B testing, monitoring and drift detection
• Effective cross-functional collaboration with store operations to refine thresholds and checkout UX timing