Owned end-to-end development of customer-facing TrustAuth and Seller Trust features for Amazon’s mobile shopping experience, deployed across 10+ marketplaces. Built scalable frontend components and backend services to surface merchant credibility signals via internal APIs, while preserving backward compatibility and experiment safety in a distributed production environment. Ran A/B experiments, collaborated cross-functionally, and independently resolved cross-platform production issues during on-call rotations.
Built an end-to-end Android application to evaluate real-world robotaxi ride experiences, owning system design, implementation, and iteration. Designed the app architecture using Kotlin and Jetpack Compose, integrated BLE devices to collect IMU sensor data during rides, and connected third-party map APIs to support accurate location tracking. The application was used in continuous test rides and informed downstream analysis of autonomous driving performance.
Designed and implemented a data evaluation pipeline to validate delivery promise accuracy at scale, establishing a baseline for ongoing promise system re-architecture. Built a proof-of-concept using AWS Glue, S3, and event-driven Lambda services to ingest, transform, and reconcile large datasets, reliably validating millions of client records per run. Defined accuracy and consistency metrics and visualized results through dashboards to analyze tradeoffs between correctness, latency, and system complexity in a production environment.
A web-based application that enables Northeastern CS students to search and explore course data and participate in anonymous, course-scoped forum discussions on instructors and course reviews
A project that visualizes the conditioning signals in ControlNet models to better understand how different inputs influence the generated outputs in diffusion models.
Built a Spark–HBase nearest neighbor fraud detection system deployed on AWS EC2, using autoencoder-compressed features and Z-order row keys to support efficient neighbor scans at scale. Evaluated system performance and scalability by varying EC2 cluster size and partitioning strategies.
A Java CLI application that manages drivers, passengers, and ride orders, implementing DAO-based persistence with MySQL and order-matching logic based on routes and availability.
Currently exploring AI-powered tools and applications. Interested in building systems that solve real problems with elegant architectures.