Sahil Yadav

Working on ad auction floor-price prediction using XGBoost and contextual-bandit models across major OEMs, targeting higher revenue with a game theory approach to understand DSP behavior.

Designed an end-to-end warehouse-store vehicle scheduling and routing model with MILP optimization under OSRM, capacity, manpower and timing constraints, achieving INR 5 Cr monthly cost reduction nationwide.

Engineered a large-scale demand forecasting engine with custom loss functions, delivering a 5% MAPE / 10% Bias reduction and a ₹22 Cr reduction in fortnightly procurement.

Led cross-functional development of Zepto's Auto-Replenishment System, an omnichannel transfers generator, yielding a 10% availability uplift and faster forecast-to-execution. Built a pick-path optimization system via co-occurrence-based placement and affinity-driven sequencing, cutting average travel distance by 30%.

Formulated a PID control-based dynamic pricing model balancing real-time demand against supply stress while minimizing order loss.

Built a relevance ranking model over video metadata using embedding similarity and NLP to improve ad conversion.

Contributed to CTLearn: trained RNN-based models for gamma-ray event reconstruction and classification, and integrated ROOT data pipelines for MAGIC and VERITAS telescope data.

Pursued an integrated dual degree program in Mathematics and Electrical & Electronics Engineering, with a focus on applied mathematics and machine learning.