• Conducted model compression research for Large Language Models (LLM) with focus on quantization and structured matrices pruning • Applied techniques from numerical linear algebra research to improve state‑of‑the‑art quantization and pruning algorithms • Designed downstream evaluation pipelines for real‑world production scenarios. • Applied the research to up‑to‑date LLaMA models and internal models, achieving ∼13% compression with <1% drop in downstream accuracy

• Improved the precision and efficiency of the Collision Detector in C++ • Built a visualization system for various safety events in C++ and JavaScript

Developed machine learning algorithms for remote analysis and monitoring of chemical experiments through measurement devices

• Developed and fine‑tuned TextCNN model for embedding product titles and queries • Built end‑to‑end ML pipelines: data generation, model training, deployment, and monitoring using Go and Python. • Cleaned product titles using BERT‑based entity extraction • Developed an easily maintainable machine learning pipeline which extracts features, refines ranking and boost targeted products during query search by XGBOOST and LR, improved the overall Click Through Rate by 15% • Applied graph theory and approximate nearest neighbor search to enable similar product recommendations.

• Unified image processing modules with C++/Python bindings • Studied effects of Gamma Correction and Color Correction on existing RestNet‑style models for road‑edge boundaries detection • Optimized image processing parameters for the deep learning models • Improved generalization of the model by implementing customized data augmentation tools through mathematical optimization • Facilitated a seamless migration of Computer Vision models from the outdated camera platform to the new one

Developed Computer Vision algorithms for indoor navigation in shopping malls