About Me

AI Engineer at Daiichi Sankyo

As an AI Engineer Intern, I designed and deployed enterprise-ready retrieval-augmented generation (RAG) systems to make large-scale knowledge retrieval faster and more reliable. I fine-tuned large language models using QLoRA and applied quantization for optimized inference, reducing memory usage while maintaining performance. By integrating Hybrid Graph RAG, I improved the contextual accuracy of responses and enabled efficient real-time analysis of medical and research documents.

I engineered a FAISS-based vector search pipeline with GPU acceleration, MMR retrieval, and graph traversal (NetworkX). This architecture improved retrieval diversity and lowered response latency, ensuring the system could scale to enterprise workloads. The solution was deployed using Flask on AWS with automated monitoring and error handling, making the pipeline both production-ready and cost-efficient.

Beyond LLM optimization, I built a data integration application that ingested medical records daily from the VVRIM application via APIs, transformed them, and stored structured outputs in Amazon Redshift. Using CRON scheduling, the pipeline automated ingestion and transformation while preserving historical consistency by recreating past datasets for any given date. This capability allowed researchers and analysts to query both real-time and historical views of the data, improving reliability, traceability, and decision-making across the organization.

Research and Teaching Assistant at UB

As a Research Assistant, I focus on improving the sustainability of AI systems by optimizing the energy consumption of deep learning workloads. I build frameworks in Python and C++ to precisely measure memory usage, execution time, and compute overhead. This research helps uncover patterns of inefficiency in large-scale training and inference, with the goal of advancing green computing practices in AI.

My work involves detailed energy and performance profiling of resource-intensive applications using advanced techniques such as multiprocessing, parallel computing, and Linux-based profiling tools. By leveraging PyTorch, TensorFlow, and low-level system monitoring, I design experiments that measure trade-offs between energy cost and performance, producing insights that guide the design of more efficient AI architectures.

In addition to research, I serve as a Grader for a graduate-level Machine Learning course, where I assess assignments, review projects, and provide constructive feedback to students. I assist in clarifying challenging concepts, guiding students on algorithms, and ensuring they grasp key principles of machine learning. This dual role enables me to apply my technical expertise while fostering an environment of learning and collaboration.

Software Development Engineer at Experian

As a Software Development Engineer at Experian, I worked extensively on the Experian Access application and contributed to the development of the CPD Waterfall solution, both of which are enterprise-scale platforms critical for secure processing of sensitive financial and PII data. My responsibilities spanned across full-stack development, backend architecture, and API integrations, ensuring these systems remained reliable, scalable, and secure.

In the CPD Waterfall project, I developed front-end components using JSP, ExtJS, and JavaScript to create an intuitive workflow for uploading PII data and generating credit reports. On the back-end, I engineered Spring and Hibernate based services with layered architectures including Controllers, Services, and DAOs to manage secure data transactions. I also optimized MySQL schemas for higher throughput, improving database reliability and performance under heavy load.

For the Experian Access application, I helped modernize and migrate its technology stack by contributing to the migration from Spring 4 to Spring 5, ensuring compatibility, improved performance, and stronger security compliance. In addition, I worked on resolving legacy security vulnerabilities and reducing technical debt, which improved system maintainability. Alongside this, I collaborated on the enhancement of an internal automation tool that leveraged Robot Framework and Selenium, refining automated testing workflows and improving development cycles across multiple teams.

These combined efforts not only supported core product innovation at Experian but also ensured that critical enterprise systems were modernized, secure, and capable of handling growing workloads with reliability. My role allowed me to bridge software engineering best practices with enterprise system modernization, making meaningful contributions to both day-to-day operations and long-term scalability.

SDE Internship at Experian

As a Software Development Engineer Intern, I contributed to the CS-Access application, a key system for credit score management. I worked on backend enhancements using Spring Boot and Hibernate, optimizing service and DAO layers for more reliable data access and smoother workflows. I also expanded REST APIs to improve interoperability with other internal systems.

To streamline repetitive tasks, I implemented CRON jobs and Quartz Scheduler, automating credit score updates and external API integrations. These automated schedules included robust logging and error-handling mechanisms, ensuring stable execution and reducing reliance on manual operations.

In parallel, I developed automated testing frameworks using Python Robot Framework and Selenium WebDriver, which expanded regression coverage and improved testing reliability. Additionally, during a company hackathon, I built a Transactions Application using Spring Boot and Hibernate to give production teams quick access to subcode and transaction data. This tool simplified day-to-day workflows and significantly accelerated ticket resolution for operations teams.

Research Assistant at KMIT

During my research internship at the Research and Development Lab of Keshav Memorial Institute of Technology, I actively contributed to advancing medical imaging analysis, focusing on breast carcinoma pathology. I developed and refined AI-based deep learning models to detect tubules in whole-slide images, enhancing the accuracy of medical diagnostics. Leveraging frameworks like Faster R-CNN and YOLOv5, I designed a robust pipeline to identify mitosis in digitized pathology images, significantly aiding in breast cancer detection. I demonstrated expertise in utilizing tools such as QuPath to construct systematic workflows for annotating Whole Slide Imaging (WSI) TIFF files, producing high-quality images and masks for accurate labeling. Collaborating within a four-member team, I effectively handled image preprocessing, implemented distributed training strategies, and showcased strong leadership and presentation skills. These efforts highlighted my ability to tackle complex challenges in the intersection of AI and medical research.