Google LLC
Senior Software Engineer
- Led end-to-end system design and execution for multiple next-generation Cloud TPU New Product Introductions (including TPU7 and TPU v5p), delivering compute engine stack turnup 3 months ahead of hardware availability and pulling in the development schedule by 4 weeks.
- Co-led a year-long cross-organizational initiative (10 teams across 5 departments) to architect and deliver a distributed Cloud TPU Pod Simulator, accelerating cloud hardware development lifecycles by 6+ months prior to silicon availability.
- Directed ML compiler and runtime (libtpu) SDK integration for tier-1 AI/LLM research customers; designed and delivered a runtime SDK within 2 months that enabled accelerator sub-slicing for large-scale inference workloads.
- Increased ML model training performance by 30% on Cloud TPU instances by resolving virtualization (vIOMMU) degradation, implementing static hugepages, and optimizing multi-PCIe root and NUMA configurations to eliminate non-local memory access.
- Drove alignment across networking, kernel, and ML framework teams to resolve systemic performance bottlenecks, establish disaggregated serving network benchmarks, and streamline the JAX stable stack and PyPI release pipeline.
