National Laboratory of the Rockies (NREL)
Researcher, Residential Buildings Group
Led production machine learning for the DOE Section 179D Web Calculator, owning data engineering, model development, validation, deployment and monitoring. Also worked on LLM agents, reinforcement learning, fault detection and building control.
- Led development of production ML surrogates for the DOE Section 179D Web Calculator, a nationally deployed, IRS-supported platform; models achieve 1.1–3.3% average CVRMSE across 16 U.S. climate zones.
- Trained ridge regression and deep neural network surrogates on EnergyPlus simulations to estimate energy-cost savings across 2 commercial building types, 6 HVAC system types and 16 climate zones.
- Built a LangGraph and LLM agentic workflow to automate grey-box RC model identification, validated by deterministic criteria: RMSE ≤ 0.5°C, parameter CV ≤ 0.15 and a winning-candidate margin above 0.1°C.
- Operationalized MLOps and observability with MLflow, DVC, Evidently, Prometheus, Grafana and GitHub Actions, plus AWS SageMaker and Azure ML, covering versioning, drift detection and monitoring.
- Co-developed a two-stage deep reinforcement learning framework for grid-interactive multi-zone building control, combining evolution strategies with PPO and benchmarking against MPC baselines.
- Created an HVAC fault-detection method using thermostat drive-cycle data and Extended Kalman Filters, published in Energy and Buildings (2021), with sensitivity-analysis findings in Applied Thermal Engineering (2024).
- Python
- PyTorch
- LangGraph
- EnergyPlus
- MLflow
- DVC
- Evidently
- Prometheus
- Grafana
- GitHub Actions
- AWS SageMaker
- Azure ML
Earlier Postdoctoral Researcher, Residential Buildings Group · Jan 2019 – May 2020