● Developed a graph-based RAG system to enhance document retrieval and response generation for enterprise chatbot app● Integrated multiple intent-based documents into a structured knowledge graph, enabling efficient data retrieval based on user’s intent.● Designed an intent-routing mechanism that guides customer service agents with contextualized information, improving response accuracy and support efficiency.● Utilized Chroma Vector DB for optimized vector storage and retrieval within the RAG system.

● Development of the core Kaya platform, focusing on implementing Python-based web services using Flask.● Developed and integrated components in a generative AI–RAG ecosystem, using libraries like transformers and LangChain.● Worked with Chroma Vector DB for efficient vector-based data storage and retrieval.● Implemented and maintained SQL-based solutions for data storage and processing.● Explored beginner-level front-end development using JavaScript to support integration with backend services.

● Used MMDetection and Tracking algorithms like DeepSORT to make a real time object tracking model● Trained off-the-shelf DL models with publicly available data using Tensorflow and Pytorch● Identified and measured baseline identity-tracking quality metrics that illustrate model performance

● Worked on the improvement of Quantitative Risk models to increase the model’s coverage beyond the traditional payment instruments from 40% to 99.8%● Developed an E2E model that has was able to achieve a 98% accuracy and 0.98 AUC● Used AWS Glue for ETL for data dumping and optimized glue jobs to decrease the DPU hours by 300%● Used SageMaker Pipelines to develop an E2E pipeline to preprocess the data, train the model, get inference, do a bias check and make an explainability report● Used SageMaker batch transform to train the model and do inference on 2TB of data● Conducted experiments to choose the best type of model from Deep Learning methods, XGBoost, GBM etc.

● Conducted a comprehensive review of different outlier and drift detection techniques for streaming data● Created a modularized framework to implement data drift detection using KL divergence, JS divergence and PSI metrics● Developed a novel drift window size optimization algorithm using Bayesian optimization and implemented multiprocessing,achieving 80% reduction in algorithm run time● Used LOF, Hierarchical Clustering, and Half-Spaced Trees for Outlier detection in real-time

● Successfully implemented key machine learning models for Insurance, Credit Card, and Commercial Real Estate from SAS to Apache Spark, ensuring a 100% match between Python outputs and SAS during validation for a seamless end-to-end migration● Scaled the ML models to cater to a massive customer base, increasing from 1 million to 250 million customers, and rigorously stress-tested for reliability● Developed a Hive monitoring automation framework integrated into post- processing modules, scalable across various consumer decision support models● Extensively used SQL, Pandas, Numpy, and Python to create utilities like datetime handlers and excel writers for efficient data processing● Honored with the "Manager’s Spotlight Award" for outstanding efforts and exceptional achievements in data science projects.● Authored bash scripts tailored for Apache Hadoop operations and innovated Python functionalities by implementing decorators for multiprocessing, multi- threading, and precise time and memory logging

● Made single-cell RNA sequencing pipeline in R and used dimensionality reduction algorithms like UMAP, t-SNE and PCA to remove noise● Made use of clustering algorithms like k-means, DBSCAN and hierarchical cluster- ing to classify the data into appropriate cell types● Wrote bash scripts to run jobs in the Yale’s Cluster

● Implemented multiple Ensemble Learning and, Deep Learning algorithms in Keras and Tensorflow to classify Cardiac Arrhythmias using UCI dataset and MIT-BIH dataset. Achieved an accuracy of 99%● Implemented various Image Processing algorithms to digitise ECG Scans using OpenCV, Numpy and Pandas● Our project "Dhadkan" was shortlisted for the Commonwealth Healthcare Awards 2019

● Improvement of Patient View Page of cBioPortal using Javascript and IGV Library.● Developed the display of Variant Track of cBioPortal using JavaScript