I turn research into models that ship.
Applied ML engineer working on retrieval, evaluation and low-latency inference. I take ideas from a paper to production traffic without losing the science.
180ms
p95 inference latency in production
12M
daily requests served by my systems
3
first-author workshop papers
Selected projects
Retrieval gateway for a support assistant
Designed a hybrid retrieval layer (BM25 plus dense) with a reranker, feeding a grounded answer model for a customer support product.
Cut hallucinated answers by 41% and raised deflection rate from 28% to 52%.
Offline evaluation harness for LLM releases
Built a reproducible eval harness scoring every model candidate on a curated suite before it reaches traffic, with regression gates in CI.
Blocked 9 quality regressions pre-release and made model swaps a one-day decision.
Low-latency inference service
Rewrote a batched inference service with continuous batching and quantized weights to hold latency under load.
Held p95 under 200ms at 12M requests per day while cutting GPU cost 34%.
Skills
Modeling
Serving
Data & Eval
Platform
Experience
-
Since 2022
Senior ML Engineer
Aperture AI
Own the retrieval and evaluation stack for a support assistant used by millions; lead a squad of four.
-
2020 to 2022
Machine Learning Engineer
Loop Research
Shipped the first production recommendation model and the team's model serving platform.
-
2018 to 2020
Research Engineer
University Vision Lab
Published on efficient inference and maintained the lab's training infrastructure.
Publications
- Continuous Batching for Low-Latency LLM Serving
- A Reproducible Evaluation Harness for RAG Systems
- Distilling Rerankers for Edge Inference
Certifications & talks
Efficient LLM Serving at Scale
MLSys Workshop, talk
2024AWS Machine Learning, Specialty
Amazon Web Services
2023Deep Learning Specialization
DeepLearning.AI
2021Let's build something that ships
Open to senior and staff ML roles, and to research collaborations on retrieval and evaluation.