As part of the core recommendations team, my work sits at the intersection of Applied Research and ML Engineering to enable the next generation of AI models and algorithms that power recommendations and personalized discovery. I collaborate deeply with Research Scientists, and partner with our ML Platform & Serving Systems teams to enable the next generation of our ML infrastructure & capabilities to improve the member discovery experience.

My primary area of work is around building & optimizing our training and data infrastructure for building Ranking models that power title/entity ranking across various product surfaces. Lately I've building engineering systems to finetune and evaluate large scale Gen AI based models for ranking tasks.

Worked on News & Feeds Ranking and Recommendations infrastructure for Microsoft Start (start.com). As part of the Ranking Infrastructure Team, I was a tech lead and contributor to various ranking infrastructure components that power personalized content recommendation for millions of Windows and Microsoft Edge users.

I was one of the founding engineers on the team who created Tally, a real-time event processing and aggregation system that was responsible for processing, storage and serving of optimized data for machine learning use-cases with low-latency reads & high-throughput writes. The system enabled us to create machine learning models trained on novel user-engagement signals that were processed & served using Tally to better understand our users and provide an improved personalized feed experience. In terms of impact, the system was responsible for 50% of top 25 most important features of our ranking models and was computing and serving critical real-time engagement data streams and ML features. My contributions across several projects resulted in significant improvements in top line engagement & revenue metrics. Additionally the optimizations I made to storage schemas, and serving components resulted in 50% reduction in CPU & memory footprint compared to v0 of the system.

As a Research Assistant at the Emotive Computing Lab led by Dr. Sidney D'Mello, I contributed to active research on modeling collaborative problem solving discourse using ML and Language Modeling. I also worked on modeling bias in machine learning models for apparent personality prediction in one way behavioral interviews.

Technologies used: Python, Pytorch, Keras

Designed and built re‑usable end‑to‑end components to operationalize solutions for automated document classification and information extraction. Reduced the turnaround time to build a new Proof‑of‑Concept for unseen set of documents from over 1 month to 1 week. The components I built included all parts of the stack from UI, Backend, Storage, Model Training, Inference, Feature Parsing and Serving

Built Stride's core frontend framework with custom components and styles using Angular that was used in products like ACE, Know-Your-Customer (KYC), and other projects, serving as a frontend boilerplate for any web based product at Stride.

Developed interactive visualizations in D3.js supporting real-time updates for timeline data, graphs of connected entities, and document clustering results.

Built a custom PDF Viewer using PDF.js which allowed users to provide feedback on results given by Machine Learning Models using annotations. The viewer also supports multiple document comparison, dual monitor support, and is being used in 10+ projects. I played a key role in memory optimization, lazy loading and cross-browser compatibility for the viewer allowing it to seamlessly handle documents with 500+ pages and reduced latency of loading 100 page pdfs from 5+ seconds to < 1 second while delivering a rich and smooth user experience.

Conducted over 40+ technical phone screens and 10+ in-person interviews for SWE internship roles, as well as full-time positions for Frontend and NLP Engineers. I was also a contributor to the internal developer handbook and authored several technical articles on Stride's internal Confluence.

Worked on 2 projects with the Cloud DevRel Team as a DPE Intern at Google San Francisco. One called Github Issue tracker which is an alternative to notification emails sent by GitHub, and another which is an internal tool to integrate GitHub issues with Google's internal issue tracker.

GitHub Issue Tracker is released as an open source project on GitHub. I worked on both projects from scratch and developed both backend servers (in Go) and a web frontend written using Angular 4 and hosted using Google App Engine.

Technologies Used: App Engine, Go, Angular 4, Cloud SQL and Firebase along with internal tools at Google.