Data scientist who cares why a model works, not just
whether it does.
I hold an M.S. in Data Science from UVA (August 2026) and a B.S. in
Applied Mathematics from James Madison University. Lately I've been deep
in agent tooling: fine-tuning an LLM with LoRA to make one correct tool
call instead of hallucinating one, writing MCP servers that reconcile
real inference cost, and building local-only NLP pipelines that
transcribe, summarize, and graph what recurs across hours of video.
Outside of data science, I breakdance — there's something satisfying
about both finding patterns in datasets and nailing a new move.
python pytorch transformers peft/lora mcp langchain spacy pyspark neo4j ollama sql fastapi git azure
LLM Cost Ledger · python, mcp
MCP server that answers what an LLM call actually cost, reconciling
ccusage-captured usage logs from Claude Code, Codex, OpenCode, and Amp.
Local-only: no proxy, no cloud infrastructure, no telemetry.
Podscope · python, pyspark, neo4j
Local NLP pipeline that transcribes YouTube videos, segments them by
topic, summarizes each two ways, extracts entities, and graphs what
recurs across a channel. No paid APIs: faster-whisper and a local Ollama
model do the work a hosted API usually would.
Frequency-Based Speech Isolation for Keyword Spotting · python, pytorch
CNN vs. Audio Spectrogram Transformer for keyword spotting under noisy
conditions, tested across three preprocessing strategies and five SNR
levels. The CNN won every configuration (93.42% clean, 81.94% at -5 dB
with bandpass filtering).
Fraud Detection ML Pipeline · python, azure
Trained on 6.3M+ PaySim transactions, 94% recall and 0.99 ROC-AUC on a
severely imbalanced dataset (0.13% fraud). Shared train/serve pipeline on
Azure ML and Container Apps, so the deployed model always matches what
was validated in training.
AURA-ED · python, llm
Clinical decision-support tool that synthesizes fragmented ED patient data
into structured early risk profiles, built on the Stanford MC-MED dataset
(118,385 adult ED visits).
Oregon Trail Fitness · fastapi, healthkit
Built at HoosHack 2026. An Oregon Trail-themed iOS fitness tracker pulling
real step data from Apple HealthKit. I built the FastAPI backend and the
HealthKit integration.
