Balaji S Sarath P.

• Achieved over a 10% increase in click-through rate (CTR) by developing and deploying a two-tower recommendation model to personalize deal offers. Followed an agile development approach and validated performance through rigorous A/B testing. • Collaborated with cross-functional teams across multiple time zones to enhance promotional messaging efficiency, achieving an estimated savings of ~5,000 man-hours through the adoption of Generative AI for content creation, optimization, and personalization. • Enhanced and automated the product data setup process by developing context-aware image labeling pipelines. • Partnered with engineering, product, and merchandising teams to align machine learning deliverables with core business KPIs, ensuring measurable impact and visibility at the executive level.

Data driven design strategies for Organic Photovoltaic devices: • Near real time microstructure generation through GANs that obeys additional constraints(material parameters, performance metrics) including those from the dataset. • Graph Neural Networks for Molecular Property Prediction Systematic exploration of GNN models and feature specification for molecular property prediction. • Developed data efficient deep neural network training strategies that can reduce high fidelity data dependance unto 99% through low-fidelity multi-sensor data • Highly efficient Finite Element solvers for Excitonic drift diffusion equation that scales well over 1024 compute nodes. • Coordinated with researchers and mentored students across several universities to create the state-of-the-art tools for ML in material science and chemistry. • Developed Quantum computing based search strategies for performance optimized microstructure in OPVs

Computational methods to engineer process-structure-property relationships in organic electronics: The case of organic photovoltaics • Highly efficient and scalable Finite Element solvers for modeling phase separation during solution processed OPV manufacturing. • Explainable Convolutional Neural Networks for OPV Performance Prediction Develop a robust interpretable CNN framework for current prediction of active layer morphology. • Data Driven Process Optimization of OPVs Develop and deploy tools (PARyOpt) for identifying champion processing conditions for organic photovoltaic devices (OPVs). • Adaptive Design of Experiments Deploy PARyOpt for automated exploration of processing space * Developed GRATE for automating analysis of HRTEM images Uses graph networks algorithms to find connectivities, short-long range order in TEM images.

Operations research (Minor)