Research & Impact.
Building on past research in deep learning and computer vision, my work is shifting toward AI decision systems for sustainability and operational challenges — leveraging forecasting, uncertainty modeling, and decision optimization.
1 items Vehicle Classification in Video Using Deep Learning
Real-time vehicle classification in traffic surveillance suffers from occlusion, variable lighting, and multi-class imbalance — limiting automated traffic management systems.
Designed a deep convolutional neural network architecture trained on augmented surveillance datasets, applying transfer learning and fine-tuning to handle class imbalance.
Demonstrated that deep learning models can achieve robust multi-class vehicle classification under realistic surveillance conditions, outperforming traditional feature-engineered baselines.
Published at MLDM 2019; established foundational experience in applied deep learning and computer vision for real-world operational systems.
Directly informs future work in AI decision systems: the same pipeline design — sensor data → feature extraction → model → operational decision — transfers to forecasting and optimization in sustainability contexts.
2 items tuneR
CreatorHyperparameter tuning for sparse Partial Least Squares (sPLS) models in mixOmics was computationally expensive and lacked reproducible defaults.
Developed intelligent statistical heuristics replacing exhaustive grid search, with deterministic seeding for full reproducibility.
On the repository benchmark harness, random search cut median wall time by 60.5% while matching the best observed accuracy across a 125-combination block-sPLSDA search space.
Demonstrates ability to bridge statistical methods with software engineering — a core skill for building reproducible decision-support pipelines.
mixOmicsIO
CreatorMulti-omics studies at scale are bottlenecked by data conversion between SummarizedExperiment objects and mixOmics matrices.
Built a memory-optimized data pipeline using reference semantics and strict S4 validation for type-safe multi-gigabyte transformations.
Simplified the data handoff between Bioconductor structures and mixOmics workflows for larger computational biology analyses.
Demonstrates production-minded data pipeline design — the same architecture discipline applies to sensor-to-model pipelines in sustainability operations.
Decision-making under uncertainty
How can forecasting models quantify operational uncertainty to improve resource allocation in dynamic environments?
Forecasting for sustainability operations
Time-series and remote sensing approaches to predict operational demand, resource availability, and environmental risk.
Reinforcement learning & operations research
Applying RL and optimization methods to sequential decision problems under sustainability and resource constraints.
AI decision systems for agriculture
Climate-resilient and resource-efficient agricultural operations as a high-impact domain for applied decision science.
Reproducible decision-support pipelines
End-to-end systems from sensor data ingestion through model inference to operational KPI reporting — built for auditability and real-world deployment.
Resume & Academic CV
View and download the Industry Resume or full Academic CV.