Sarthak Sharma

About Sarthak Sharma

I’m a Senior Software Engineer at Oracle with 7+ years building and scaling multi-tenant AI services. I lead cross-functional teams to design distributed ML platforms and harden large-scale GPU infrastructure, shipping secure, compliant features fast. My toolkit spans Python/Java services, cloud architecture, and MLOps—and I obsess over developer experience, reliability, and cost. Recently helped scale training to 1,800+ GPUs across 5 regions and lift utilization by ~60%.

Experience

1. Senior Software Engineer

   Oracle

   Sep 2022 — Present · Seattle, WA, US

   Built secure multi-tenant data/model management and interactive sessions for OCI AI Studio; integrated and hardened SSH; standardized NFS (and Lustre where needed) to balance throughput with DX.Architected and launched Octane, a multi-tenant, multi-region distributed training platform for large-scale models.Scaled a production fleet to 1,800+ GPUs across 5 regions; lifted utilization ~60% via driver/image pinning, phased canary rollouts, and scheduling improvements.Tuned RDMA/RoCEv2 (PFC/ECN/QoS) and NCCL collectives for scale-out; added topology-aware configs and NUMA/GPU affinity.Enforced image/driver consistency (NVIDIA drivers, Fabric Manager, OFED) across heterogeneous clusters; built an OL8/OL9 (UEK & RHCK) Ansible imaging pipeline.Shipped NVML-based GPU health/telemetry and preflight validations to quarantine bad nodes before jobs.Delivered end-to-end observability (Prometheus/Grafana/ELK), metered cross-tenancy billing, and usage reporting—improving unit economics and cost transparency.Authored benchmarking suites, “golden-path” templates, runbooks, and reference architectures—accelerating experiment velocity and reliability for DS/ML teams.Established a Center of Excellence for GPU efficiency (sizing guides, RDMA/NCCL tuning playbooks, failure-mode library) and influenced OCI service roadmaps for performance & safety.Introduced a lightweight quality-review mechanism that sped releases to the OCI Marketplace without compromising security, compliance, or operational readiness.

Skills

• Public Speaking
• C++
• Matlab
• Pl/Sql
• R
• Data Analysis
• Cascading Style Sheets (Css)
• Python
• Machine Learning
• Microsoft Powerpoint
• C
• Programming
• Sql
• Linux
• Git
• Peoplesoft
• Computer Networking
• Microsoft Excel
• Peopletools
• Microsoft Word
• Mysql
• Microsoft Office
• Html
• Cloud Computing
• Peoplecode
• Javascript
• Java