Master of Science (M.S.) in Computer Science

San Diego, California

Bachelor of Technology (B.Tech.) in Computer Science

Vellore, India

IIT Bombay

Indiana University - Purdue University

• Built and optimized ETL pipelines ingesting and pre-processing large-scale member, claims, and pharmacy data (20M+ records/month) using PySpark and Airflow, improving data processing throughput by over 40%.
• Developed an agentic AI system leveraging LangChain/LangGraph, Hugging Face, and open-source LLMs (LLaMA 2, Mistral) to automate clinical summarization, streamline member triage, and enhance knowledge retrieval, saving 10+ hours per week.
• Hide
• Hide

• Architected the core marketplace platform using T3 Stack(Next, TypeScript), that supported 10+ partner orgs, 50+ projects and 200+ student users.
• ⁠Designed and enforced least-privilege access for critical operations with comprehensive audit logging using NextAuth.js and RBAC via CASL, cutting incident investigation time by 20%.
• Built an asynchronous orchestration layer for LLM workloads, scaling long-running inference tasks (extraction, spec generation, recommendations) to handle 3× higher throughput without blocking core services.
• Designed an LLM-based data processing pipeline that extracts key project signals from unstructured requirements into structured specs, saving 20 engineering hours per week.

• Optimized the Lenders & Advisors analytics platform using Apache Spark and AWS to deliver real-time insights on real estate listings, expediting updates by 30%.
• Automated Data pipelines using Apache Airflow, facilitating the transfer of structured and unstructured data into RiVirtual’s centralized data lake on AWS S3 for streamlined analysis and reporting.
• Built a recommendation system using collaborative filtering in PyTorch to deliver tailored listings and property suggestions to users.
• Built a custom dashboard for Realtors, with Tableau to track engagement of users.

• Designed and built the PAIR AI fashion recommendation Model and the vision processing pipeline using OpenCV, TensorFlow, and Deep Neural Network Models for enabling precise fashion pairing and product recommendations.
• Architected backend systems and multi-layered REST APIs using Google Cloud Services and Firebase for managing aggregated user data and deploying AI Models.
• Implemented a secure user auth and session management system using OAuth 2.0, JWT tokens in Google Identity Platform.
• Designed Wireframes on Figma and consequently developed and maintained the iOS and Android App-PAIR Virtual Stylist.

• Designed and implemented a Decentralized Authentication with Device-to-Device Trust using Elliptic Curve Diffie-Hellman protocol, expediting authentication and reducing overhead by 28%.
• Designed and Built the Voice Command Module using Amazon Alexa Speech API, created and tested command routines for voice assisted control of appliances and console.
• Optimized API call volume by batching calls using gRPC and cached responses using Redis, reducing command response latency by 32%.
• Fixed a major bug that would log users out and clear history after a firmware update.

• Funding: Foundation of China (Grant No.61471228), The Key Project of Guangdong Province Science & Technology Plan (Grant No. 2015B020233018) and the Scientific Research Grant of Shantou University, China, Grant No: NTF17016.
• Developed a novel approach for classifying vasculature in digital medical imaging using Zernike Moment-based feature vectors with neighborhood Gaussian kernels.
• Designed supervised pattern recognition models utilizing moment-invariant feature vectors as learning parameters.
• Led research initiatives, including data curation, formal analysis, methodology development, project administration, software implementation, visualization, and review/editing of technical reports.

• Funding: NUS School of Computing
• Optimized Feature Pyramid Networks (FPN) for detection tasks, enhancing Single Shot Multibox Detector (SSD) accuracy for specific scenarios.
• Conducted an in-depth study of ResNet-50, adapting reinforcement learning techniques for analyzing recurring video footage, improving detection consistency
• Implemented time-multiplexed thread parallelization to reuse last-layer outputs, increasing detection accuracy by 8% and reducing computational overhead by 20%. Developed a 3D scene prediction model to forecast object positions 2–3 seconds into the future using time-multiplexed multi-instance position and momentum data.

• Funding: VIT - School of Computer Science and Engineering (SCOPE)
• Conducted research and development of a standalone navigation aid for visually impaired individuals.
• Conceptualized the architecture of the device, leveraging Darknet YOLO for real-time object detection and Python libraries like OpenCV for vision processing.
• Integrated hardware components, including the NVIDIA Jetson Nano, camera, and microphone, using Linux-based workflows and NVIDIA CUDNN for optimized neural network performance. Developed a standalone system capable of object detection and proximity estimation through real-time image processing, eliminating the need for network connectivity or additional sensors.