I am a proud alumnus of Georgia Tech, where I completed my Master's in Computational Data Science, and IIT Kharagpur, where I earned my Bachelor's in Mechanical Engineering. My academic journey through these prestigious institutions has ingrained in me a profound understanding of data science and technology, blending theoretical knowledge with practical application.
I have recently embarked on an exciting journey as a Senior Data Scientist at Prudential Financial in their Chief Data Office. My extensive background in Data Science encompasses domains like Recommender Systems, NLP, Graph Machine Learning, and Statistical modeling. Prior to this, my role as a Senior Data Associate at Innovaccer Inc. involved developing innovative solutions in healthcare analytics. My internship at Prudential Financial was a pivotal experience, enhancing my skills in Graph machine learning and NLP.
Beyond the professional realm, I am an avid sports enthusiast, enjoying Soccer, Tennis, Cricket, and more. I have a deep fascination with global cultures, driven by an eagerness to explore different countries and their unique stories. Additionally, I am a keen follower of tech and entrepreneurial podcasts, constantly seeking insights into the art of value creation in business.
Financial reports offer critical insights into a company's operations, but require costly manual review. To address this, we leveraged finetuned Large Language Models (LLMs) to distill key indicators and operational metrics from these reports basis questions from the user. We leverage the FinQA dataset to fine-tune both Llama-2 7B and T5 models for customized question answering. We achieved results comparable to baseline on the final numerical answer, a competitive accuracy in numerical reasoning and calculation.
For FinTech, this offers a tool to quickly interpret complex financial documents, aiding in rapid and informed decision-making. It can enhance real-time analysis capabilities for investors, credit analysts, and regulatory compliance officers, providing a competitive edge in the fast-paced financial sector
This project explores enhancing Large Language Models' (LLMs) reasoning and question-answering capabilities by integrating Chain-of-Thought (CoT) reasoning with Visual Question Answering (VQA) techniques. Utilizing TextVQA and ScienceQA datasets, the research assesses the effectiveness of combining text and visual embedding methods to improve LLMs, focusing on solving multiple-choice questions and enhancing reasoning abilities.
Implementing this multimodal reasoning framework in the retail and e-commerce sector can significantly enhance customer search experience and product discoverability. By effectively integrating text and visual data, this approach offers more accurate and contextually relevant search results, driving sales and improving customer engagement.
arXiv GitHub
This project demonstrates the innovative application of Large Language Models (LLMs) for generating synthetic medical and pharmacy claims data. The approach centers around converting tabular data into text prompts and fine-tuning LLMs to recreate realistic data patterns. This method addresses the limitations of conventional generative models by capturing time dependencies and complex relationships within healthcare data.
Utilizing Large Language Models for synthetic data generation provides a strategic advantage across various enterprises. It ensures data privacy and compliance, enriches data for robust analysis, and supports innovative solutions in fields like finance, retail, and healthcare. This technology facilitates deeper insights and decision-making while protecting sensitive information.
Details GitHub
RoboChef merges image recognition with personalized recommendations, utilizing CNNs with ResNet models for food classification and a recommendation engine based on collaborative filtering and matrix factorization. It uses the Food-101 and Food.com datasets for accurate and user-tailored suggestions, optimizing with methods like SVD and NMF. Performance metrics include error rate and accuracy, improved by advanced training techniques.
RoboChef offers a scalable solution to enhance customer experience in the food and health sectors by providing personalized meal suggestions. It takes into account individual preferences and dietary restrictions, which could lead to increased customer satisfaction and loyalty. The adaptability of the system to various user constraints makes it a valuable tool for businesses aiming to cater to the personalized nutrition market.
Details GitHub
This project report encapsulates the creation of a hybrid music recommendation system using Spotify's API and MLHD data, involving data extraction, machine learning algorithms, and a Tableau dashboard to deliver finely-tuned, user-centric music suggestions.
The hybrid recommendation approach, blending user behavior with content attributes, provides a more nuanced and accurate recommendation, crucial for platforms seeking to enhance user engagement and satisfaction. The explainability aspect of recommendations fosters user trust and transparency, a key differentiator in competitive markets. Additionally, the interactive dashboard can serve as a model for user-friendly interfaces, further improving user experience and increasing platform loyalty.
arXiv GitHub
The "Best Buy Project" presentation showcases a sophisticated sales forecasting model for slow-selling SKUs. The model, developed using Croston and XGBoost methods, aims to predict sales with high precision a week in advance. It addresses challenges like data sparsity and variability in sales patterns, enhancing forecasting accuracy for products with intermittent demand.
This model offers e-commerce chains significant benefits, such as optimized inventory management, reduced costs from overstocking, and improved customer satisfaction through better product availability. Its ability to forecast demand for slow-moving items supports strategic decision-making, enabling more efficient supply chain operations and potentially boosting sales through targeted marketing and pricing strategies.
Details GitHub
The report presents innovative recommender systems using the HG-GNN and ISCON methods. These approaches enhance session-based recommendations by combining current user session data with historical user behavior patterns. Tested in different domains, these methods demonstrate improved accuracy in personalizing user experiences. This advancement signifies a major leap in recommendation technology, offering more relevant and engaging content to users.
These systems have the potential to revolutionize business engagement strategies. These systems not only cater to immediate user preferences but also smartly incorporate historical data, providing a highly personalized user experience. This can dramatically boost customer retention and spending, as users are more likely to engage with content or products that resonate with their unique interests and past behaviors.
Details GitHub
