Tejas Lohia

I'm a final-year Computer Science and Engineering undergraduate at IIT Gandhinagar, with a minor in Mathematics. At Caltech's Center for Data Driven Discovery, I'm developing machine learning methods for period regression in irregular astronomical light curves.

I replaced trial-frequency grid search with direct regression on pairwise-difference statistics, reaching 93.8% of predictions within 10% relative error on difficulty-matched synthetic curves. I'm also building PeriodFormer, a frequency-aware foundation model, and a multi-GPU feature pipeline that reduces cache construction from roughly 13 days on CPU to about 20 minutes.

Previously at IIT Bombay, I worked on multi-camera 3D tracking using Kalman filtering, geometric constraints, pose estimation, and hierarchical clustering. I also built BTC and ETH trading strategies with Python, achieving Sharpe ratios of 1.97 and 2.83 in vectorised backtests.