Vedant Srinivas

Designed and tested with Nevada DOT, a low-power edge AI solution combining GPU-accelerated YOLOv5 computer vision and RF-enabled smart signage to detect and classify wildlife in real time, achieving >99% accuracy to actively warn drivers and reduce animal-vehicle collisions.

Built and deployed a computer vision model trained on simulated and real thermal imagery to automate wildlife monitoring for WSDOT's I-90 Snoqualmie Pass East Project, achieving >97% precision and recall in species classification to reduce false positives and enable scalable, real-time analysis of motion-activated thermal camera data.

Created Transformer, GRU, and 3D CNN models with multi-video finetuning for surgical phase detection. Matched state-of-the-art accuracy using 6× fewer parameters, enabling real-time, compute-efficient inference.

Engineered a modular NBA performance prediction system with automated data pipelines, API integrations, and SVD-based modeling to generate real-time stat predictions with robust validation tracking.

Built an AI-driven educational platform integrating project idea generation, mentor-matching algorithms, and customized cold email creation in React.

Improved energy efficiency of Class 8 trucks using biomimicry of a boxfish to make them more aerodynamic, published in Stanford Intersect Journal.

Designed a self-sustaining roadside device to deter wildlife using randomized audio-visual cues triggered by detecting vehicle approach via ambient light intensity. Built custom PCBs in KiCAD, programmed Arduino for real-time detection, integrated solar+battery power, and 3D-modeled the enclosure in Fusion 360 for outdoor deployment.