Progressed through multiple AWS AI/ML teams. Promoted to SDE II in April 2025.
May 2025 – Present
MLE · Custom Model Optimization (CMO)
AWS Generative AI Innovation Center
Leading R&D in LLM post-training — CPT, SFT, and Reinforcement Learning (GRPO, DPO) — for models at tens-to-hundreds of billions of parameters (Nova, Qwen families). Training infrastructure with FSDP, DDP, HuggingFace, and Lightning on SageMaker HyperPods. Production inference with vLLM, Triton Inference Server, TensorRT, ONNX Runtime, and HuggingFace TEI on EKS and SageMaker endpoints. Also driving prompt optimization and model steering research.
Aug 2024 – May 2025
SDE → SDE II · Bedrock Model Customization
Led development of Supervised Fine-Tuning for LLMs (Llama 3 family) in Amazon Bedrock. Re-architected custom model inference from merged adapters on provisioned throughput to adapter hotloading on base model capacity — enabling efficient multi-tenant serving without dedicated compute per adapter.
Apr 2024 – Aug 2024
SDE · SageMaker AutoPilot & Canvas
AutoML for tabular data, time-series forecasting, and language model fine-tuning. Helped make ML accessible to non-practitioners through Canvas's no-code interface.
Feb 2023 – Apr 2024
SDE · Managed Streaming for Apache Kafka (MSK)
Kafka on AWS Graviton chips and MSK Express Brokers with a fully managed storage layer. Optimized performance and cost for real-time streaming workloads at scale.