Developer & researcher
I build software and do research where data gets messy. Right now that means two live products, a trading screener and logistics software for trucking companies, plus research decoding EEG signals, turning biology text into pathway diagrams, and auditing clinical risk models for who they fail.
A market research desk for active traders and analysts: factor screener, backtester, watchlists and an AI analyst. The product runs a factor model over the S&P universe with a backtester behind it.
Python · FastAPI · factor models · backtesting
A 256-channel net records the scalp. Spatial filtering and a trained classifier turn microvolts into a decision: did the person just see the target or not? Pick a classifier and a stimulus, then present it.
Spatial filtering · LDA / SVM / RVM · d′ and AUC
T2PW reads plain-language descriptions of biological pathways and rebuilds them as structured diagrams. It extracts entities and reactions, enriches them, and grounds each one to HMDB, KEGG or ChEBI.
LLM extraction · ID mapping · PWML / SBML · Python
Glucose is phosphorylated by hexokinase to form glucose-6-phosphate, consuming one ATP and releasing ADP. The glucose-6-phosphate is then isomerized to fructose-6-phosphate by phosphoglucose isomerase.
Using MIMIC-III intensive-care records, this study measures how often disease-prediction models miss real cases (the false-negative rate) and whether those misses land unevenly across demographic groups. It then tests a reweighting fix across five conditions, with every change significance-tested. The honest answer is mixed: it helps in some cohorts more than others.
MIMIC-III · algorithmic fairness · significance testing · Python
The best way to reach me is email: [email protected]. Code lives on GitHub.