Soumya Mandal

AI/ML Backend Developer at Arizona State University

Developed recommender system prototypes using collaborative filtering and gradient-boosted models (XGBoost), later productionized to serve 50k+ daily users. Implemented deep learning pipelines with TensorFlow and PyTorch for personalization tasks, achieving 25% improvement in prediction accuracy. Designed A/B testing frameworks and experimentation tools to evaluate recommendation strategies, enabling data-driven deployment decisions. Optimized backend services for low-latency inference, reducing response times by 30% while scaling to thousands of concurrent users. Collaborated with designers, researchers, and engineers in Agile sprints to deliver personalized features aligned with user engagement goals.

Software Developer at Arasan Chip Systems Inc

Designed and deployed ML-driven personalization features for backend services, improving content recommendation accuracy by 22% and user engagement by 15%. Built scalable data pipelines in Python to process millions of daily records, enabling real-time recommendation and personalization services. Partnered with product managers and data scientists to test, optimize, and refine recommendation models, reducing iteration cycles by 20%.

AI/ML Backend Developer at Arizona State University

Developed recommender system prototypes using collaborative filtering and gradient-boosted models (XGBoost), later productionized to serve 50k+ daily users. Implemented deep learning pipelines with TensorFlow and PyTorch for personalization tasks, achieving 25% improvement in prediction accuracy. Designed A/B testing frameworks and experimentation tools to evaluate recommendation strategies, enabling data-driven deployment decisions. Optimized backend services for low-latency inference, reducing response times by 30% while scaling to thousands of concurrent users. Collaborated with designers, researchers, and engineers in Agile sprints to deliver personalized features aligned with user engagement goals.

Software Developer at Arasan Chip Systems Inc

Designed and deployed ML-driven personalization features for backend services, improving content recommendation accuracy by 22% and user engagement by 15%. Built scalable data pipelines in Python to process millions of daily records, enabling real-time recommendation and personalization services. Partnered with product managers and data scientists to test, optimize, and refine recommendation models, reducing iteration cycles by 20%.

Software Developer at Larsen & Toubro

Built ML-enabled microservices that powered personalization workflows for enterprise clients, reducing content delivery mismatches by 18%. Automated data ingestion and preprocessing pipelines to support training of large-scale models, reducing preparation time by 40%. Partnered with cross-functional teams to implement experimentation dashboards, improving visibility into model performance across iterations. Enhanced reliability of recommendation services by designing monitoring tools and automated retraining pipelines, reducing stale predictions by 25%. Contributed to production-quality ML code in Python and Java, adhering to best practices in testing, code reviews, and CI/CD.

Software Developer at Larsen & Toubro

Built ML-enabled microservices that powered personalization workflows for enterprise clients, reducing content delivery mismatches by 18%. Automated data ingestion and preprocessing pipelines to support training of large-scale models, reducing preparation time by 40%. Partnered with cross-functional teams to implement experimentation dashboards, improving visibility into model performance across iterations. Enhanced reliability of recommendation services by designing monitoring tools and automated retraining pipelines, reducing stale predictions by 25%. Contributed to production-quality ML code in Python and Java, adhering to best practices in testing, code reviews, and CI/CD.

Software Developer Intern at Indian Institute of Technology Guwahati

Prototyped ML models for content classification and ranking, achieving 92% accuracy in research datasets used for recommendation tasks. Supported researchers in scaling prototypes to production-ready services, reducing deployment preparation time by 20%.