PRJ-001 · GridCastActive2026 — present
Probabilistic load forecasting for campus-scale energy systems.
A 96-hour forecasting system for university microgrids. A temporal-convolutional backbone produces quantile forecasts, recalibrated with conformal prediction and validated against four years of interval data — then served through a lightweight inference API.
- Time-series
- Probabilistic ML
- Energy systems
Fig. 01 — Load forecastP50 / 90% CI
ObservedForecast90% interval
- Horizon
- 96 h
- Coverage
- 94.2%
- MAPE
- 7.8%
Student pod · Faculty advisedRepository
PRJ-002 · CarePathActive2025 — present
Simulation and optimization for outpatient care logistics.
A discrete-event model of referral routing, appointment capacity, and no-show behaviour across partner clinics — paired with a mixed-integer scheduler. Pilot scenarios reduced modelled median wait time by 18%.
- Wait reduction
- 18%
- Scenarios
- 1,200
- Sites
- 04
- Health systems
- Operations research
- Simulation
PRJ-003 · CausalTwinActive2025 — present
A reproducible engine for observational policy evaluation.
Doubly robust estimators, negative controls, and sensitivity bounds behind one scikit-learn-compatible interface — built for studies that must survive scrutiny, not just a single notebook.
- Estimators
- 12
- Validation
- Synthetic
- License
- MIT
- Causal inference
- Econometrics
- Open science
Code, tests, and a reproducibility manifest ship with every project.
GitHub organization