Capturing expert vision
Eye tracking during real inspections turns tacit expertise into data: where experts look, in what order, and why. Published in Scientific Reports, Buildings, and the Journal of Building Engineering.
Footprint
Where the network lives
Research program
Publications →
Three connected threads
Teaching machines to look
Distilling gaze-informed attention into vision transformers compressed for the edge, so UAVs can inspect with expert priorities on board — the active post-PhD direction.
Forecasting demand at scale
At GA Telesis I built a production system for forecasting intermittent, hard-to-predict demand — ensembles of temporal transformers and gradient-boosted models delivering probabilistic forecasts to decision-makers daily.
Dec 2025 — Present
Data Scientist · GA Telesis, Digital Innovation Group
- Took a demand-forecasting system from concept to production, with strong forecast coverage for intermittent demand across multi-month horizons.
- Led ensemble expansion across temporal transformers and gradient-boosted models, delivering meaningful accuracy gains over the initial baseline.
- Designed a document-AI prototype for privacy-sensitive document workflows using vision-language models, and contributed analysis that informed the broader processing strategy.
- Built the MLOps backbone: cloud ETL, experiment tracking, automated batch inference, and BI dashboards for stakeholders.
Jan 2025 — Nov 2025
Independent Researcher
- Extended the doctoral research into a working Python prototype simulating expert inspector attention patterns for training inspection models.
- Peer-reviewed computer vision manuscripts for academic journals.
- Deepened production-ML skills — AWS, MLOps, time-series forecasting — in preparation for the industry transition.
2020 — 2024
PhD · Penn State, BEAM Lab
- Developed gaze-supervised computer vision models for automated damage assessment — expert inspectors' eye-tracking data guiding CNN attention to improve generalization to unseen structures.
- Optimized deep CNN architectures for high-resolution anomaly detection: 20% lower inference latency at 85% detection accuracy for edge deployment.
- Built multimodal pipelines synchronizing eye-tracking telemetry with UAV video streams.
- Led industry-backed research projects and mentored 8 undergraduate researchers.
2018 — 2020
Graduate Researcher · Chung-Ang University
- Deployed real-time deep CNNs for automated safety monitoring on edge devices — 25% faster dynamic risk detection.
- Built the end-to-end UAV imaging pipeline with custom image-stitching over high-resolution geospatial data from 50+ inspection flights.
2018
Development Engineer · NUST
- Engineered a wireless, time-critical sensor system for fault identification — embedded C++ on edge microcontrollers.
- Achieved 97% data-transmission reliability in deployment.