Karthik Bappudi Sundar

Applied Scientist @Zalando

Berlin, DE

Multilingual Intent Recognition: Developed a model to automate customer case handling across multiple European markets, reducing manual processing of customer emails by 12%.Product Defect Detection: Created multilingual classifiers to identify product defects from customer reclamation requests, leading to corrective actions with suppliers and resulting in savings of €400K per half year.Cost-Sensitive Learning for Imbalanced Classification: Developed classifiers using cost-sensitive learning techniques to improve the detection of critical safety issues, even with a very low occurrence rate of 0.005%.LLM-Based Automated Customer Service: Designed a POC for automated customer service agents using LLMs with function calling, and established evaluation metrics to assess the effectiveness of LLMs in automating customer cases.RAG for Customer Queries: Developed and deployed a RAG system to answer customer queries. Implemented and leveraged LLM-based evaluations to ensure high-quality RAG performanceMLOps: Successfully deployed multiple machine learning applications using CI/CD pipelines. Developed dashboards and implemented monitoring systems to ensure continuous model performance and reliability.Advanced NLP and ML Techniques: Addressed cold start problems and enhanced dataset preparation through unsupervised learning, semantic search, and synthetic data generation. Applied cross-lingual knowledge transfer methods such as knowledge distillation and multilingual adapters for zero-shot cross-language performance. Fine-tuned encoder models and Supervised Fine-tuning (SFT) of LLMs using PEFT techniques, including LoRa and adapters, to optimize performance for specific tasks.

SRM IST Chennai

Bachelor’s Degree, Mechanical Engineering

Politecnico di Milano

Master’s Degree, Materials Engineering and nanotechnology