Headline

Senior Applied Scientist at Microsoft | ISI Kolkata | Jadavpur University

About

Applied Scientist with over 7 years of experience building high-impact machine learning solutions in e-commerce, healthcare, and enterprise search domains. Specialized in developing production-scale NLP systems, large language models (LLMs), and recommendation algorithms, with a proven track record of delivering solutions that have generated over $300M in revenue at Amazon and UnitedHealth Group. Skilled in Python, TensorFlow, PySpark, AWS, and Azure, with deep expertise in Transformers, BERT, GPT, and graph neural networks. Currently working as a Senior Applied Scientist at Microsoft, where I am focused on improving the M365 Copilot enterprise chat and search experience. My work involves leveraging advanced NLP and semantic search techniques to enhance productivity and information retrieval for enterprise users. Previously, I led the development of theme-based search and semantic matching systems at Amazon, driving significant improvements in customer satisfaction and revenue. At UnitedHealth Group, I designed and deployed a high-accuracy Transformer-based model for disease prediction, resulting in improved patient screening and substantial cost savings. Passionate about mentoring, technical leadership, and continuous learning, I have guided teams of applied scientists and interns, introduced standardized MLOps practices, and conducted knowledge-sharing sessions on advanced ML techniques. My research and publications span topics such as deep learning for image retrieval, multicollinearity correction, and interpretable deep learning systems.

Location

Bengaluru

AI stack

claude code

claude

gemini

hugging face

langchain

openai

Work experience

Senior Applied Scientist at Microsoft, Oct 2025 - Present · 1 yr 1 mo, Bengaluru, Karnataka, India, Working on ML solutions to improve M365 enterprise copilot chat and search.

Applied Scientist II at Amazon, Sep 2023 - Oct 2025 · 2 yrs 2 mos, Bengaluru, Karnataka, India, Theme-based Search in Amazon Fresh: Developed an advanced search recommendation system that improves search experience by grouping products in the search results into thematic categories. Utilized large language models (LLMs) for theme generation, including fine-tuning proprietary LLMs for domain-specific knowledge, and implemented Transformer encoder-based models for product assignment, theme ranking, and product ranking within themes. Incorporated exploration-exploitation strategies to balance showing established high-performing themes with discovering new potentially valuable ones. Semantic Matching in Prime Video Search: Developed and implemented an online Bi-Encoder semantic matching model for Prime Video search ranking, enhancing the accuracy of search results by accommodating semantically related queries and variations in spelling and language.

Research Scientist at Amazon, Oct 2021 - Sep 2023 · 2 yrs, Bengaluru, Karnataka, India, Substitutes in Prime Video Search: Collaborated with Prime Video and IMDb teams to onboard substitute titles for Prime Video title queries. Employed a multi-modal bi-encoder + cross-encoder and LLM-based approach, leveraging image, text, and historical behavioral metadata to compute the substitute titles. Homepage Recommendation in Prime Video: Collaborated with the Prime Video team to develop a GBDT model utilizing title-level features to assess the popularity of Prime Video titles. This model is used to rank ASINs within carousels on the Prime Video homepage, enhancing the customer experience upon landing on the webpage.

Data Scientist at UnitedHealth Group, Jul 2019 - Sep 2021 · 2 yrs 3 mos, Greater Bengaluru Area, Deep Patient: Designed, developed, and implemented a high-accuracy (97%) Transformer-based (BERT) time series model for predicting the occurrence of 70+ critical diseases. The model utilized comprehensive patient data, including medical history, lab tests, and treatment regimens. SmartRFP: Developed an innovative NLP-based solution enhancing Request for Proposal (RFP) responses in the U.S. insurance bidding process. Combined multimodal encoder (BERT + CNN) with model-agnostic approach to identify influential tokens/phrases. Data Quality: Spearheaded development of an automated NLP solution leveraging BERT for detection and masking of sensitive data in tabular datasets.

Data Science Intern at Reliance Retail, May 2018 - Jul 2018 · 3 mos, Navi Mumbai, Market Basket Analysis: Conducted Market Basket Analysis to identify item associations within key categories (Home & Personal Care, Food (Non-Staples), and Staples) for the Andhra Pradesh region. Driver Analysis: Leveraged data pre-processing techniques and Random Forest modeling to identify key drivers influencing sales across 25+ product categories for AJIO E-Commerce, providing actionable insights for business strategy.

Engineer at Bechtel Oil, Gas, & Chemicals, Jun 2016 - Jul 2017 · 1 yr 2 mos, Gurugram, Haryana, India, BP West Nile Delta: Analyzed and created automated pipelines to prepare Datasheets, MRQ and Technical Bid Evaluation using Python. Implemented basic ML algorithms (regression models and decision trees) to predict resource requirements and identify procurement patterns. Developed classification models to automatically categorize technical specifications and vendor bids based on historical project data.

Education

M.Tech, Operations Research and Applied Statistics, Indian Statistical Institute, Kolkata

Bachelor of Engineering (B.E.), Mechanical Engineering, Jadavpur University

Public links

github: https://github.com/pidahbus