Developed a large-scale firearm identity matching system by processing 3D scans of bullet toolmarks. The system is trained on the open NBTRD dataset and enhances search efficiency by transforming search time from minutes to sub-seconds. This was achieved by replacing an open-sourced R engine with a Java implementation of Hierarchical Navigable Small World (HNSW) for vector search. Utilized and adapted open-source algorithms such as CMC and CMS to create a model that translates 3D scans into searchable vector embeddings.

AI solution for customer care and sales tailored specifically for financial service providers.

Developed a robust and efficient backend for the event websites associated with popular gaming titles.

Developed a cloud gaming service that leverages kernel-based virtual machines and NVIDIA GPUs to function as game servers, with customized open-source Moonlight clients.

A start-up providing AI recommendation services for small to medium enterprises. The AI agent is capable of reading, seeing, hearing, and understanding user behavior to recommend relevant products and services. The inception of this project began with a research paper.