LLM evaluation · public code
When retrieval helps
A multi-model QA harness that separates retrieval quality from answer accuracy and tracks the cost of each question.
Projects
From experiments to useful software
Model training, careful evaluation, and interfaces that make complex systems easier to use. A selection of research, engineering and personal work.
01Software & experiments
AI interfaces · prototype
HospTrack / Nursor
A spatial hospital-workflow prototype with AI chart briefings and natural-language navigation across synthetic records.
Reinforcement learning · team project
Demonstrations to self-play
A Rocket League pipeline for behavior-cloning pretraining, PPO self-play and controlled teacher-guidance experiments.
Full-stack product · private app
Intendo
A deliberate daily planner with transactional history, revision-checked saves and an agent CLI that previews and verifies changes.
02Research & engineering
ECCV 2026 · published research
Tessellating the Earth
A location encoder with learnable spherical partitions. Co-authored research with public code and pretrained weights.
Radar engineering · Uhnder
PySCOTT
A Python/Dash modeling tool whose predictions informed production test limits and customer coverage maps.
03More work
MVRL · under review
Hierarchical image retrieval
Research and engineering for efficient image geolocalization: model training, embedding caches, budget-aware search and reproducible evaluation. Title, code and results remain withheld during review.
1088 Advisors · 2026
Research-enterprise analytics
Public funding, publication and patent data assembled into university benchmarks, embedding-based research clusters and automated reports.
Vitana · 2023
Healthcare data pipelines
Spark ETL connecting HL7v2 lab results, FHIR resources and EMR records in PostgreSQL and MongoDB, with Apache Superset for exploration.
Ozden Lab · 2019–2021
Behavioral experiment controller
A Python/C and Arduino control system with a custom PCB: 85% lower cost and 87.5% lower latency than the commercial alternative. Presented at SfN and ABRCMS 2019.
Earlier experiments & coursework
Probing LLM math
Experiments on internal representations in mathematical reasoning.
Code →GPT-2 on FineWeb
A training exercise spanning data preparation, tokenization, the training loop and evaluation.
Cloud forum & chat
Coursework deployed on AWS and GCP to explore hosting, storage and backend development.