|
I am a software engineer at Amazon Music, working on subscription growth across voice and mobile. I hold a master's degree in computer science from the University of Illinois Urbana-Champaign. I am broadly interested in full stack software development, distributed and networked systems. Outside of work, you can find me on a road bike, camera in hand, on the hunt for a new coffee shop I can read a book at. Email / Resume / LinkedIn / Github / Photography |
|
|
|
|
Software Development Engineer @
Sep 2024 - present | Seattle, WA Amazon Music Growth: Tier Progressions Built and shipped experimental upsell experiences delivered through proactive voice notifications at scale. Also worked on Android initiatives for upsells and engagement. Together these have helped drive 150k+ annualized Amazon Music tier signups. |
|
Graduate Teaching Assistant: Distributed Systems @
Jan 2024 - May 2024 | Champaign, IL Worked with Prof. Radhika Mittal on the course CS425: Distributed Systems - asynchronous clocks, global system snapshots, failure detection, concurrency control, consensus via paxos/raft, leader election, distributed transactions, DHTs and P2P systems etc, distributed datastores etc. Mentored students to build distributed & networked application projects on reliable multicasting, distributed transactions, raft, distributed file system. |
|
Software Engineer Intern @
May 2023 - Nov 2023 | San Jose, CA Designed and developed a new core feature for ONE-DL, the company's primary product, that moves network data from on-premise environments to the cloud via Kafka. Built a network packet collector that taps high-throughput data-center traffic to extract and stream metadata, using Kafka and Elasticsearch for pipelining. Added a REST API and Redis cache for selective packet capture and data sink integrations. Optimized the pipeline with payload batching and DPDK-based kernel bypass strategies, achieving 3x throughput and 50% lower latency. Benchmarked with Scapy and Ixia load generators. |
|
Full Stack Developer Intern @
Nov 2020 - Dec 2020 | Bangalore, India Developed UI with ReactJS and wrote REST API endpoints with node.js and Express.js for two different MERN stack applications. Integrated custom Tensorflow object detection models with Shinobi (open-source CCTV framework). |
|
|
|
Designed a distributed job scheduler system for ML inference tasks built on top of 10 Linux VMs from scratch using Java and Python. Uses a real-time work scheduling algorithm to optimize query rate for ResNet and ImageNet classification tasks. Includes a distributed data logging service, distributed group membership protocol and failure detector, a distributed files system. |
|
Page Fault Profiler: A kernel-space virtual page fault profiler, monitoring page faults and CPU run times via a virtual memory buffer and delayed work queue at 20 samples per second, and managed process registration through the proc interface. Rate Monotonic Scheduler: A real-time CPU scheduler as a Linux kernel module based on Rate-Monotonic Scheduling (RMS) for single-core processors, utilizing the Proc filesystem for user space communication. Computing CPU Time: A Linux kernel module for process management using a doubly linked list and proc filesystem, featuring periodic CPU time updates via a workqueue and timer callbacks. |
|
Best Ethereum Powered Project @ DevSpace '21 by CSI, VIT & Best Dyte Powered Project @ DevSpace '21 by CSI, VIT A fundraising platform for wild life sanctuaries that allows for donations via ETH. There are live-streams from high activity sites at the sanctuaries that allow audience to tune in, watch and donate for a cause. Demo youtube. |
|
MERN Stack business intelligence web application that works as a website/catalogue for a yarn trading entity in Erode, TN, India to receive business inquiries for products. Data collected is used to create visualizations for business intelligence. Also designed and implemented an alternative JavaFX frontend. |
|
A MapReduce algorithm for finding frequency of unique words a large volume of text tested on a single machine with multiple docker containers acting as separate nodes for map and reduce jobs. The throughput of the task is checked by scaling the data and nodes up and down. |
|
A collection of projects around interactive computer graphics - Implementations of a Rasterizer and Ray Tracer from scratch that take descriptions of a scene and output graphic images. Also includes a collection of work with WebGL: creating a 3D Model of an arbitrary landscape terrain using random fault generation, illumination via diffuse/specular lighting models and erosion simulation. A flight simulation of said generated terrain. Finally a simulation of particles with physical forces like momentum, gravity and drag. For more details and demonstrations of this work. |
|
In the this project, a comprehensive benchmarking study was conducted on the complexity of real-world bugs in open-source Maven and Gradle projects. Popular bug datasets such as Defects4J, BugSwarm, Bears, and QuixBugs were utilized to evaluate the quality of test suites generated by automated test generation tools, specifically Randoop and Evosuite. Code statement coverage for these bugs was analyzed using tools like Clover, IntelliJ IDEA, and JaCoCo. Suspiciousness scores for code statements were calculated from the coverage reports to identify lines of code with the highest likelihood of containing bugs. Results youtube. |
|
OLA: Implemented OLA to progressively obtain approximate query results for various Pandas DataFrame operations, such as filtered mean, grouped mean, and grouped sum. Apache Arrow-Parquet Encoder: Implemented HyperLogLog to estimate the dataset cardinality. Modified the Apache Arrow source code to implement a new ASCII Encoder and Decoder for Parquet, supporting the Integer and Float data types. Flatbuffers & Shared Memory: Worked with Google's Flatbuffers and Python's shared memory libraries to pass serialized DataFrames between notebook sessions, and performed various operations directly on the serialized DataFrames. This improves data sharing and manipulation efficiency in Python since leveraging Flatbuffers provides faster serialization and shared memory for inter-process communication. |
|
|
|
International Conference on Artificial Intelligence and Machine Vision (AIMV), 2021 Presented a comparative analysis of benchmarks for latest security frameworks from then recent IoT literature while advised by Dr. Anil Kumar Kakelli. Categorically classified and critiqued existing IoT security frameworks based on their approaches to address the threat of malignant nodes in heterogenous device networks and general strengths/limitations. |
|
design and source code from Jon Barron's website |