A collection of my research contributions and real-world engineering work, from published papers to hands-on systems and open-source projects.
My research spanning federated learning, encryption, recommender systems, distributed systems, and much more.
We propose HERL, a reinforcement learning-based approach to dynamically optimize homomorphic encryption parameters across client tiers in federated learning. HERL improves utility by 17%, reduces convergence time by up to 24%, and increases efficiency by up to 30% while maintaining security.
Read it here →We propose a novel iterative deep unfolding algorithm to remove fixed-pattern stripe noise from infrared images. By leveraging neighboring column correlations, our method preserves fine details and achieves real-time performance.
Read it here →A comprehensive overview of recommendation system types, methodologies, limitations, and industry applications. This work has been cited ~500 times and remains one of the most downloaded in its domain, with ~30,000 downloads.
Read it here →We present BOLT, a hyperconverged design for Docker registries that improves scalability, caching efficiency, and deployment simplicity by consolidating registry components into a tightly connected cluster.
Read it here →We investigate techniques to mitigate heterogeneity challenges in federated learning across diverse client devices and datasets, enabling more performant and equitable collaborative model training.
Read it here →My practical engineering work, research prototypes, and open-source contributions demonstrating applied problem solving in AI, distributed systems, and software development.
A Python-based tool that parses LaTeX research papers and automatically extracts metadata such as publication year, equations, references, and section structure for downstream analytics and knowledge graph construction.
View on GitHub →A retrieval-augmented generation (RAG) system that uses FastAPI, SentenceTransformers, and FAISS to ingest documentation and resolve support tickets automatically. Designed for scalability and local LLM deployment with Ollama.
View on GitHub →A full-stack recommendation engine that takes a user’s resume and returns the top five most relevant job listings using LDA topic modelling, TF-IDF similarity metrics, and a React + Django web app. The system was trained on a corpus of 20,000 job postings and designed to combat information overload in job searches.
Read the documentation here →An iterative deep learning framework that progressively removes fixed-pattern stripe noise from infrared images while preserving fine detail, leveraging unfolding principles to reduce search space and improve convergence.
View on GitHub →Contributed to the open-source JupyterLab-Git plugin by improving user credential caching, significantly enhancing user experience during remote repository operations.