Deepak Yadav

Experience

1. Machine Learning Data Engineer

   Dentsply Sirona

   Mar 2023 — Present · Waltham, MA, US

   Developed PIE-Net, a PointCNN-based deep learning pipeline for automated 3D margin line detection on 16,384-point intraoral scan point clouds, predicting per-point confidence scores and directional offset vectors with engineered geometric features (surface normals, curvature spikes) to localize sharp crown-to-tooth transitions.• Resolved extreme class imbalance (<5% margin points) using Focal Loss (alpha=0.95, gamma=2.0), achieving recall of 0.961 and precision of 0.851, and built a post-processing pipeline using Dijkstra path-finding, skeletonization, and Taubin smoothing to produce clean closed 3D curves from raw model output.• Achieved 30% clinical expert acceptance rate, more than doubling the industry baseline, with end-to-end inference time of ~4.3 seconds per scan deployed via Flask/gRPC into the DS Core restorative workflow.• Architected and deployed a large-scale ML inference platform on GCP, standardizing embedding artifact lifecycle, model versioning, and multi-environment CI/CD release pipelines across 6+ multi-tenant ML services on 10 regional production clusters, cutting per-service onboarding from weeks to days.• Built distributed ML inference runtime with gRPC API sidecars, PostgreSQL result storage, and NVIDIA T4/L4 GPU nodes with Kubernetes AutoScaling (HPA/VPA), processing 530+ jobs/day at zero error rate with p95 latency under 6 minutes across all production services.• Owned Kubernetes cluster management across 10 regional clusters using Terraform and Kustomize IaC, orchestrating multi-tenant workload isolation, AutoScaling policies, and multi-environment deployments with automated smoke-test CronJobs executing live end-to-end inference validation on every release.• Drove ML observability and SLA reliability across 3,730+ weekly inference jobs via Datadog APM, standardizing latency, throughput, and success rate metrics, enabling proactive alerting and rapid incident resolution across 8 production market regions.

Education

• Northeastern University

  Masters in Data Analytics, Information Technology