jmd.gg

About

Now, and before

I am a machine learning engineer at Amazon AGI in Sunnyvale. I designed and built Reinforcement Fine-Tuning for Amazon Nova: the customization path that lets enterprise teams train frontier models on their own data, on SageMaker HyperPod. It shipped in the AWS re:Invent 2025 CEO keynote. Day to day that means SFT, DPO, PPO, and GRPO, plus the stacks underneath them — NeMo, VERL, vLLM, PyTorch Lightning.

Before AGI I was on Route 53, where I owned ML-powered threat intelligence for the resolver fleet: detectors fast enough to sit inline on the query path, heavier async models for signature work, and the plumbing that pushes every verdict across a global fleet in real time. That fleet answers trillions of DNS queries a day, which is a good place to learn that "probably fine" is not a number. Along the way I shipped DNS Firewall features, contributed to Route 53 Profiles, and automated away about 1,200 engineer-hours of fleet work a month.

The apps are where I test my own opinions about products. Suelto, PostureMaxx, and hey, did you are live on the App Store; Proof is still on the bench. All of them are small, opinionated, and built to survive contact with a normal person's Tuesday.

The writing is how I find out what I actually think. Every essay here started as an argument with myself about what models should remember, what they should retrieve, and when good enough should ship.

  1. Apr 2025 to present · Sunnyvale

    Machine Learning Engineer

    Amazon AGI

    Reinforcement Fine-Tuning for Amazon Nova, used across SageMaker and Bedrock. Led a team of three through delivery. Announced at the Re:Invent 2025 CEO keynote, opening custom frontier-model training on HyperPod. Work across SFT, DPO, PPO, GRPO, plus NeMo, VERL, vLLM, and Lightning.

  2. Aug 2022 to Mar 2025

    Software Development Engineer II

    Amazon, AWS · Route 53

    Owned threat intelligence for the resolver dataplane. Inline and async models, realtime distribution across a global fleet, DNS Firewall, Profiles, and automation that cut about 1,200 engineer-hours a month. Led five engineers. Ran 40+ L4-L5 interviews.

  3. May 2022 to Aug 2022 · Washington, D.C.

    Software Engineer, Mobile

    TRED

    Built the mobile app from scratch on a team of three. React Native, AWS, Firebase, Stripe, GraphQL, Redux.

  4. 2016 to 2022 · Washington, D.C.

    Information Technology Consultant

    Office IT Solutions

    Technology for 15+ clients: web and application development, plus the marketing stack around it. Led a team of three consultants.

Past businesses

2019 to 2021

Urbu

A DTC brand that reached 10k customers in 10 countries and grew 50% year over year before a small acquisition.

2019 to 2020

Desktrak

Racetrack silhouette sculptures for about 15k customers and a set of track partners, with JIT inventory manufactured in China.

B.S. Computer Science, University of Maryland, 2021. A.S. Computer Science, magna cum laude, Frederick Community College, 2015.