// TRANSMISSION INBOUND
Aviation nerd turned data scientist. I build machine learning systems, data pipelines, and the occasional robot.
I started out obsessed with aircraft and ended up obsessed with the data they generate — and everything else worth measuring. These days I work across the stack: exploratory analysis, machine learning models, and the orchestration that keeps data moving in production.
Recent work includes deep-learning detectors for AI-generated imagery, LLM-based multi-agent systems, and Airflow environments automating real administrative workloads. I like problems where research meets deployment — where the model has to actually survive contact with the real world.
A fine-tuned detector for AI-generated imagery that explains itself: every verdict ships as a JSON breakdown across 8 criteria for why an image reads Real or AI. Runs on-device — validated on Snapdragon X Elite, Snapdragon Gen 5, and a Rubik Pi 3's Adreno GPU — or via Hugging Face and llama.cpp.
view repo →A quadruped robot crawler wired up as an interactive assistant — perception, control, and conversational commands running on a Raspberry Pi.
view repo →A DIY 'sentry mode' for vehicles that never shipped with one. Object detection on a mobile compute unit acting as a dashcam watchdog for non-IoT cars.
view repo →A dockerized MLflow environment (MinIO, Postgres) that simulates drift in a live pipeline and triggers the right response — full retrain, doubly-robust reweighting, or incremental SGD updates. An experiment in drift-resistant pipelines and algorithmic bias.
view repo →A live Airflow environment automating recurring administrative and data tasks at Purdue. DAG design, scheduling, and the unglamorous reliability work that makes automation stick.
🔒 private — Purdue infrastructureOrchestrated migration of data from S3 into Redshift with Glue, plus downstream visualization. Includes the scripts and queries used to validate every stage.
view repo →A processing pipeline orchestrated by Airflow and deployed via Docker, moving data through S3, Spark, and Hadoop end to end.
view repo →A bidirectional LSTM forecasting weekly unemployment claims across Indiana census tracts, mapped with choropleths and heatmaps. Took 2nd place in Purdue's IronHacks data science competition.
🔒 private — competition entryQuantitative comparison of airspace density over major US airports before and during COVID-19 — measuring exactly how hard the pandemic grounded air travel.
view repo →A DMAIC overhaul of a Purdue Honors College process that demanded manual review every semester. Mapped the key phases, isolated the bottlenecks, and cut lead time from 107 days to 21.
🔒 case study — available on request// published work, and flight plans still being flown.
A systematic scoping review of AI in healthcare supply chains — PRISMA screening, bibliometric analysis, and LDA topic modeling. Co-author. Sustainability 2026, 18(3), 1434.
LLM agents that can plan, retrieve, and verify — automating real administrative workflows in a university setting.