Projects

Projects

Selected reproducible research and evaluation projects by Jiayi Deng, including the Human-AI Fairness Audit Lab.

Selected public work focused on reproducible evaluation, psychometric evidence, behavioral data, and responsible AI assessment. The featured project is a synthetic, non-hiring audit demonstration designed to make fairness and workflow risks inspectable.

Featured

A public, reproducible audit of a synthetic AI-assisted candidate-review workflow. The project evaluates held-out model performance, subgroup and intersectional outcomes, calibration, counterfactual behavior, reviewer overrides, escalation policy behavior, and monitoring regressions.

Responsible-use boundary: synthetic applicants and simulated decisions only. This is an evaluation demonstration, not a hiring recommendation system.

  • Python
  • scikit-learn
  • Fairlearn
  • DuckDB
  • Streamlit
  • pytest
Training cases
6,000 synthetic
Validation cases
2,000 synthetic
Held-out audit
2,400 synthetic
Monitoring
800 synthetic

Selected research and software work

`aberrance` R package

Open-source R package work for detecting aberrant behavior in test data, connected to peer-reviewed research in Applied Psychological Measurement.

See selected research

Rapid guessing and effort-moderated scoring

Simulation and meta-analytic research on response-time thresholds, rapid guessing, reliability, test performance, and population invariance in equating.

View publication archive

AI effects evidence synthesis

Current systematic review and meta-analysis work on AI's role in student motivation and engagement, listed as under review in the supplied content source.

Read research overview