Vaidyanath Shanthakumar, PhD

• Innovate, design, implement, deploy, and maintain core Machine Learning algorithms and highly scalable, low-latency, high-throughput distributed Machine Learning systems leveraging on-premise and cloud infrastructure driving Product Discovery and Personalization experiences for multi-millions of users and items on Bed Bath & Beyond, Overstock.com, and Zulily.
• Offline evaluation of models and validation by designing and establishing correlation with online A/B tests.
• Designed a quick and efficient strategy to implement Shop-the-Room complementary recommender system to identify products in style images and recommend similar looking products in the product catalog using Foundation Models and LLMs.
• Designed and developed a real-time, in-session personalization recommender system.
• Designed an efficient strategy and suggested multiple measures to allow users to filter similar item recommendations that reduced user bounce rate by 12%.
• Fine-tuned a Llama3 based multimodal model to improve its accuracy on a product attribute identification task.
• Designed, developed, and deployed a cost-effective end-to-end embeddings based recommender system for similar item recommendations that replaced the current ensemble of algorithms that increased the CTR from 2% to 4%.
• Played a key role in the efforts to identify and correct partner entered product attributes using Generative AI.
• Developed an automated ETL pipeline for daily image data processing using Google Cloud Composer, improving ML project timelines by 60% and enhancing iteration speed.
• Successfully mentored interns through research projects, and supported and assisted in the design and development of an in-house embedding analysis and visualization tool for ML scientists.
• Reporting and Publishing novel approaches invented through peer reviewed papers.

05/2020 - 08/2020

• Innovated novel strategies to overcome data sparsity issues and successfully aggregated various types of user and item embeddings using a novel multi-modal fusion approach to successfully predict and aggregate missing modalities.
• These embeddings outperformed the existing complex ensemble of similar item recommendations algorithms in production and increased the CTR by 3%.
• Designed and implemented a novel method to approximate Matrix Factorization output that increased the coverage by 1000% and with a reasonable quality of recommendations as compared to the Matrix Factorization approach saving significant computation time.
• Published technical papers in peer reviewed conferences. Won the best paper award.

• Designed and developed a novel job recommender system in collaboration with the United States Department of Justice (DOJ).
• Designed and developed a song recommender system with the Million Song Dataset using Graph Mining techniques.
• Used Natural Language Processing techniques to analyze user reviews for deep sentiment analysis with the Yelp dataset.
• Worked for the US Army Research Laboratory towards a research project under the US-India Defense Technology and Trade Initiative (DTTI)
• Project - Image categorization interface in Virtual Reality - Designed and developed a interface with multimodal interaction capabilities including natural hand gesture recognition, speech recognition, and eye gaze for enhanced natural human-computer interaction in Virtual Reality using Oculus VR with the Unity 3D Game Engine. Also, conducted multiple user studies as a part of user-interface/user-experience (UI/UX) research.

• Deep learning based time series regression for signal processing to predict the location of a transmitter in GPS-denied environments.
• Built a novel end-to-end model that outperformed the state-of-the-art ML-based approach in literature both in accuracy as well as computation efficiency.

• Introduction to C Programming (CS 102) - (instructor)
• Introduction to Video Game Design and Programming (CS
  347)
• Introduction to Computer Graphics (CS 445/545).

Responsibilities: Designing syllabus. Designing and grading homework, assignments, projects. Mentoring undergraduate students.