Latham & Watkins
Senior Machine Learning Engineer
Building enterprise search and contract intelligence platforms from the ground up, including document parsing, structured extraction, citation verification, and human review workflows.
Experience
Palantir Technologies
Deployment Strategist
Built and deployed AI systems for mission-critical government and defense applications, including an LLM-based workforce classification platform and real-time radar/camera sensor fusion for drone tracking.
SMX
Machine Learning Engineer
Built LLM extraction and translation services deployed on Kubernetes/OpenShift in secure government environments. Developed validated, monitored data pipelines to support ongoing model retraining.
Deloitte Consulting
Senior Consultant
Processed COVID-19 lab testing data for the CDC Emergency Operations Center and built data products used in briefings to CDC leadership and the White House Task Force. Co-led a 200+ member Spark community, developing reusable tools and training analytics practitioners.
Selected Projects & Open Source
Healthcare Navigator · CareNav
View repository ↗Built a healthcare question-answering application centered on grounded responses, privacy safeguards, and human escalation. Answers require source citations; the system also redacts protected health information and supports handoff to a person when needed.
Requests are routed across model cost tiers to balance inference cost and answer quality. An evaluation suite checks response quality and provides a way to assess reliability as the application evolves.
Audition With Me
An AI-powered rehearsal partner for actors. The application turns scripts into interactive scenes with voice-powered scene partners, so performers can practice their lines independently.
As a major product and engineering contributor, built guided rehearsal experiences with role selection, voice casting, live speech tracking, and teleprompter playback. Developed parsing for typed, scanned, bilingual, and annotated scripts; integrated contextual ElevenLabs voice delivery, caching, and persistent casting; and led a cinematic, responsive interface redesign.
Podcast Search Engine · In progress
Building a search engine for more than 800,000 podcast episodes. The retrieval pipeline combines keyword and vector search to find candidates, then reranks results to improve relevance.
A two-tower model learns query and episode representations using hard-negative examples to distinguish closely related results. ONNX-based serving targets end-to-end latency under 500 ms.
GEPA in DSpyNet
View repository ↗Implemented GEPA in DSpyNet, the open-source .NET port of Stanford's DSPy framework. The contribution adds feedback-driven prompt optimization to help language-model programs improve their instructions through evaluation rather than manual prompt editing alone.
The optimization process uses feedback from model outputs to generate and refine candidate prompts, extending the framework's tools for iterative LLM development.
Background
- Education
- M.S. Computer Science — Artificial Intelligence, Georgia Tech
B.S. Bioengineering & B.A. Economics, University of Maryland - Focus
- Retrieval & ranking · Embeddings & two-tower models · Applied LLMs, RAG & agents · Classification & extraction · Evaluation · Training pipelines & low-latency serving
- Tools
- Python · PyTorch · Go · C# · TypeScript · Spark · Airflow · Kubernetes/OpenShift · ONNX · pgvector · DSPy/GEPA
Interests
I'm interested in production AI, information retrieval, distributed systems, sensing, and intelligent physical systems—especially turning emerging research into systems people can use and depend on.