San Francisco Bay Area
Applied ML Systems Lead with 15+ years of experience owning end-to-end technical decisions. I shape applied ML decision-making in the physical AI space at Amazon, guiding how models are evaluated, selected, tuned, and deployed into production environments. My work sits at the intersection of applied machine learning and systems engineering. I specialize in translating ML models into reliable systems under real-world constraints such as data quality, latency, and cost, and in navigating the tradeoffs that determine whether ML succeeds in production. I am most effective in ambiguous, high-leverage environments where technical judgment directly shapes product direction and long-term strategy.
Applied ML & AI Systems Owned applied ML decisions in physical AI space, establishing evaluation and selection criteria for vision and vision-language models under real-world constraints. Built benchmarking pipelines enabling consistent model selection across perception workflows (defect detection, classification, device/UI recognition). Productionized vision and VLM models (e.g., Grounding-DINO, Qwen-VL, YOLO, RT-DETR) under tight latency targets (10s–100s ms), single-GPU constraints, and varying data sizes (hundreds to 100k+ samples), prioritizing generalization and rapid adaptation to new products and defect types. Improved detection F1 from ~0.70 to >0.99, reduced sim-to-real gap from ~18% to <3%, and achieved <0.5% escape across thousands of runs. Designed an adaptation agent for VLMs that optimized inference configuration, reducing onboarding time for new objects from days to under an hour (initiative highlighted in Andy Jassy’s AgentCore GA announcement). Authored technical tenets guiding evaluation and integration practices for perception and agentic workflows. Developed a grounded NL-to-SQL analytics assistant (Aurora + Redshift) that creates queries, reasons over results, and provides visual insights; authored a technical paper outlining architecture and evaluation. Systems & Platform Leadership Architected a validation automation and data platform adopted across 7+ Devices teams and external partners. Built distributed AWS pipelines and edge services supporting 10+ devices across 60+ production tests. Introduced a low-code platform enabling stakeholders to build operational tools. Led delivery of manufacturing platforms for Amazon Kuiper terminals, identifying critical pre-launch issues. Leadership & Organizational Impact Served as technical lead across cross-functional initiatives spanning hardware, software, operations, and applied ML. Hired an engineering team (100+ interviews), promoted engineers, converted interns to full-time.
Led redesign and performance optimization of 5+ production test platforms supporting high-speed optical modules (100G–400G), improving throughput and reliability across multiple product lines. Led framework-level software optimizations that reduced production test cycle time by up to 20% across a ~$20M test program, improving capital efficiency as the platform scaled to additional product lines. Played a key technical role in high-volume optical R&D programs, integrating hardware validation, automation systems, and data analytics across distributed test infrastructure
Architected and built an end-to-end production automation system from work order through assembly, calibration, testing, and analytics for 40G QSFP+ manufacturing. Designed core backend architecture (multi-tier data layer, parallel execution engine, process flow control, configuration management). Wrote ~60% of the platform codebase in a team of four, establishing technical direction and execution standards. Enabled scalable manufacturing workflows supporting thousands of devices per week that were shipped to top hyperscalers.
Developed signal processing algorithms (deconvolution, filtering, frequency correction, aging compensation) for optical channel monitoring systems. Built calibration and validation software to ensure measurement accuracy and long-term device stability. Bridged research concepts and production deployment for optical communication systems.
Conducted research in integrated photonics and nanoscale opto-electrical sensing. Designed experimental systems merging photonics, microfluidics, and nanopore devices. Applied signal processing and modeling techniques to single-molecule detection systems.