Sparsh Naik

I build real-world AI systems focused on decision-making and perception under uncertainty, with a strong emphasis on translating models into deployable, reliable pipelines.

In my projects, I’m focused on end-to-end performance and robustness: I developed a Vision-Language-Action (VLA) system for robotics that maps visual input to action decisions using vision-language models and control logic, while prioritizing latency and real-world constraints. I also designed reinforcement learning for dynamic pricing under uncertain demand, inventory constraints, and competing sellers—implementing and comparing SAC, PPO, DDPG, and contextual bandit baselines in a shared simulation framework.

I’m equally comfortable working at the edge of ML and systems. I built an IoT + edge analytics smart helmet for accident detection using multi-sensor fusion (accelerometer, gyroscope, GPS), and I developed an end-to-end ML modeling pipeline for the WiDS Datathon, achieving an F1 score of 0.78 and a global rank of 17. Alongside building systems, I lead applied learning through the Data Science Club as President and contribute to cloud/DevOps technical sessions on the Cloud and DevOps Club core team.