Oct 2023 – Present · 2 yrs 8 mos

• Gen AI Infrastructure Leadership:
• Direct the delivery of critical Gen AI capabilities, including LoRA support and Voice Agents. Collaborate with LLM serving optimization teams to unblock new revenue streams by enabling customers to deploy thousands of custom models in Bedrock and SageMaker AI with minimal TTFT while only paying for compute actually used. Architected robust detection and recovery mechanisms for critical workloads across the latest NVIDIA GPUs as well as Amazon Trainium / Inferentia accelerators. Engineered the platform to ensure high reliability for models served using vLLM, DJL, and Triton.
• Cost & Efficiency Strategy:
• Spearheaded cost-optimization initiatives, architecting multi-model serving and smart routing strategies that maximize KV cache reuse and hardware utilization (NVIDIA/Trainium) to drive down customer TCO.
• Team Building & Talent Development:
• Manage a high-performing organization of 15 engineers. Directly hired 6 key members and mentored 4 engineers to promotion (Junior to Mid/Senior levels), fostering a culture of continuous growth and high talent density.
• Strategic Customer Impact:
• Partner with Product, Science, and Technical Account Managers to define the product roadmap. Directly manage relationships with strategic enterprise customers representing $10M+ in annual revenue, resolving critical escalations and driving platform enhancements that prevented churn and secured long-term adoption.
• Operational Excellence & Culture:
• Transformed the team’s operational culture by instituting rigorous COE (Correction of Error) processes and leading weekly operations reviews. Established mechanisms to identify repeated failures and drive long-term architectural fixes, maintaining 99.99% availability for a platform serving 3 million+ requests/second.

Oct 2021 – Oct 2023 · 2 yrs

• Technical Leadership: Led a team of 8 engineers to build core time-series forecasting functionality within SageMaker Canvas and Autopilot, democratizing ML for low-code/no-code users.
• Performance Engineering: Re-architected data processing pipelines using Spark and Pandas to remove blocking dependencies, delivering 50% faster training and 50% faster inference than existing solutions with zero loss in accuracy.
• Service Ownership: Led the Amazon Forecast service, enabling customers to automate heavy lifting tasks like data featurization, hyper-parameter tuning, and back-testing without requiring ML expertise.
• Customer Impact: Enabled major enterprise customers like Foxconn to improve forecasting accuracy and successfully handle supply chain disruptions driven by demand fluctuations during the Covid pandemic.

Jan 2020 – Jan 2022 · 2 yrs