Samyak Deshpande

• Designed and developed an edge-based gun detection system from scratch to beta release, collaborating with cross-functional teams to transition from server-based to real-time edge video analytics, a first for the company.
• Optimized deep learning models (YOLOv7/v8) through OpenVINO IR conversion and INT8 quantization, achieving 30% latency reduction and 91% throughput improvement (36.73 → 70.26 FPS).
• Engineered an automated video processing pipeline integrating object tracking, preprocessing, and model inference using Intel DLStreamer.
• Developed a Lua-based event streaming system, enabling low-latency transmission of gun detection and scene metadata from edge devices to the cloud, improving event processing efficiency.

• Contributed in ML projects, including predicting Customer Lifetime Value in Retail and identifying creditworthy borrowers.
• Authored 14 articles on core machine learning topics, enhancing knowledge of Decision Boundary and Large Counts Condition.
• Conducted research on Label Smoothing techniques and Interpretable ML approaches, resulting in invitation as co-author for upcoming technical publication.

A FastAPI application that allows users to create and manage event triggers that fire based on specific conditions. Includes Testing support and password authentication.

Developed ChronosAPI, an event-triggering platform using FastAPI, Celery, Redis, Docker, and JWT authentication. It supports asynchronous processing, real-time trigger testing, and has full Swagger documentation.