robustness-testing

AI-detector auditing toolkit: measures how often a detector flags genuine human writing, how stable its verdict is across seeds, and whether that verdict survives meaning-preserving edits. Detector-in-the-loop measurement harness. Claude Code skill + Python CLI. MIT.

Compositional robustness testing for vision models

Is your RAG understanding your data - or guessing?

End-to-End Python replication of Camara & Aublin's (2025) monetary spillover analysis methodology. Implements rotational-angle decomposition, Bayesian VAR with Normal-Wishart priors, sign restrictions for shock identification, and a full robustness suite for international macroeconomic analysis.

Multimodal deepfake detection with explainable AI, robustness validation, and calibrated trust scoring for real-world media.

Parkinson's disease classification from PPMI voice and clinical features with SHAP explainability, noise-robustness testing, and UPDRS progression modeling.

End-to-End Python implementation of Semantic Divergence Metrics (SDM) for LLM hallucination detection. Uses ensemble paraphrasing, joint embedding clustering, and information-theoretic measures (JSD, KL divergence, Wasserstein distance) to quantify prompt-response semantic consistency. Based on Halperin (2025).

Fine-tune with TEA

Systematic USTEC/Nasdaq CFD research archive with educational MQL5, Python tooling, transparent backtests and multi-year robustness analysis.

Robust spatiotemporal traffic forecasting on METR-LA & PEMS-BAY with GRU + graph-temporal models, strict time splits, robustness tests, and conformal uncertainty.

Evidence-first infrastructure for quant research, event-driven backtesting, robustness analysis, and trading RL.

Reproducible robustness and safety evaluation for low-cost SO-101 robot policies.

Reproducible quant research: failed strategies, backtesting probes, execution-cost studies, and engine defects found with RLXBT.

End-to-End quantitative (Python) decision support system for optimizing economic resilience against disasters. Implements updated MRIA model using multi-regional supply-use tables, three-step optimization algorithm, and comprehensive impact assessment to identify vulnerabilities from production concentration and logistical constraints.

A modular pipeline that audits 5 small HuggingFace LLMs across transparency, fairness, robustness, explainability, and privacy. Produces per-pillar scores across the 5 pillars of AI trustworthiness.

Truncated Hisotry Framework for Synthetic Control Approaches Code

A local-first Python framework for fuzz testing computer-vision models with configurable video transformations.

A reproducible edge-case audit of Merlin CT image-text behavior.

Machine learning pipeline for kidney stone risk prediction, featuring calibrated models, interpretability (Permutation Importance + PDPs), and a clean modular architecture for clinical decision support.

RobustnessPilot — Automated LLM evaluation framework testing 3 models (14B–70B params) across 23 failure modes with 7 prompt strategies | First-author paper submitted to IEEE SRDS 2026