Experience

Rochester, New York

• Built a Bayesian continual-learning framework that dynamically adapts network depth and width for evolving tasks; published at ICML 2024.
• Developed a cross-task representation-alignment framework that improved average accuracy by 2.16 percentage points for exemplar-free class-incremental learning; manuscript under review.
• Designed a parameter-efficient adaptation method for continual generalized category discovery using full-covariance Gaussian prototypes, evaluated on medical-imaging datasets.
• Developing a Bayesian adaptive graph neural network for gene–disease association prediction over protein–protein interaction graphs.

Remote

• Developed a mixture-of-experts (MoE) architecture with rank-1, LoRA-style experts for continual user modeling, enabling parameter-efficient adaptation across sequential recommendation and user-attribute prediction tasks; published at ECIR 2026.
• Designed a semantic expert-selection strategy that routes each task to relevant prior experts, reducing cross-task interference while reusing transferable knowledge.
• Outperformed the strongest baselines by 2.88% relative HitRate@5 (recommendation) and 1.37% accuracy (attribute prediction).

Kathmandu, Nepal

• Led the development of a multimodal machine-learning pipeline for identifying potential trafficking activity in online advertisements using video, image, and text data.
• Built image–text contrastive models for advertisement matching, face-based identity linking, and BERT-based social-handle extraction, improving cross-ad linkage accuracy by 35%.
• Developed a lightweight object detector that increased inference throughput by 47%, then deployed it on NVIDIA Jetson Nano devices for real-time waste-type and disposal-intent classification.
• Authored computer-vision and time-series instructional materials for the Fusemachines AI Education Program, supporting the training of 1,000+ junior engineers.

Kathmandu, Nepal

• Taught an undergraduate course covering linear algebra, calculus, probability, statistics, and information theory for machine learning.

Kathmandu, Nepal

• Worked on license-plate localization with convolutional neural networks and multiple loss functions.
• Trained to build a face-recognition system spanning detection, point-based alignment, embedding models, and nearest-neighbor classification.