bootstrap-statistics

MobileNetV2 pneumonia classifier validated on an independent 485-sample cross-operator cohort. 96.4% sensitivity, 96.4% ROC-AUC, bootstrap p=0.978. FastAPI inference API, Streamlit dashboard, DICOM support, Docker-ready

Multi-strategy crypto trading bot with Claude AI as qualitative filter. 4 parallel strategies (spot reversal, perp shorts, trend-following, funding arbitrage) on Binance. Bootstrap+DSR validation. Full 6-week paper trading case study included. DISCONTINUED — preserved as research artifact.

Measuring Semantic Drift in Multilingual LLM-Generated Clinical Trial Explanations: A Reliability Framework

Reproducible MEDAI deferral simulation (AIRI 2026). Synthetic research code.

Physics-informed simulation of projectile aerodynamics, investigating how fin-length-to-diameter (L/D) ratios influence range and velocity decay. Featuring RK4 integration, bisection optimisation, and statistical bootstrapping.

CI gate for RAG retrieval quality. Declares retrievers (TF-IDF, FAISS dense, BM25, hybrid RRF) in config, evaluates recall/MRR/nDCG with bootstrap CIs against a committed labeled dataset, and fails the build on statistically significant drops vs a git-versioned baseline. Caught a 7-point recall regression; 97% test coverage.

Streamlit app for AB test analysis using spearmint