Research Scientist at Meta (Facebook)
Menlo Park, CA | 12/2023 – Present
I work on the Marketplace Recommendation & Delivery Infrastructure team building highly-scalable data pipelines to compute, store, and retrieve machine learning features & training data, primarily in C++ and Python. Also, I implement system changes, flexible APIs, and monitoring to improve the latency and reliability of offline and online feature serving.
Software Engineer, Platform at C3 AI
Redwood City, CA | 09/2022 – 11/2023
As a member of the Platform - Data team, I worked on machine learning infrastructure problems. I researched, built and optimized the feature store, a system that stores machine learning features and serves them to models for predicting phenomena or events. My methods transform and fetch the data from the database with low latency. I analyzed the complexity of my algorithms mathematically and experimentally and identified bottlenecks to speed them up and reduce cloud computing costs on AWS and GCP. Finally, I organized meetings with data scientists and engineers to discuss my findings, analyze efficiency and investigate optimizations to reduce computation time across multiple machines.
Software Engineer Intern (Ph.D.) at Meta (Facebook)
Menlo Park, CA | 05/2022 – 08/2022
Developed multiple debugging components for machine learning feature authoring used in the data pipelines of Facebook Marketplace. The main component was a framework that categorizes errors during feature compilation, generates alerts, and assigns tasks to the appropriate team; this framework was integrated with the CI/CD. Another end product of my work was an internal UI tool to fetch and display feature values from low-latency storage after a series of transformations.
Software Engineer Intern, Platform at C3 AI
Redwood City, CA | 06/2021 – 08/2021
Implemented an end-to-end framework for cluster failure prediction; the framework has two components. The first is the data pipeline which loads cluster health metrics, handles missing data, and creates a training data set. The second component is the ML pipeline which trains a model and makes predictions regarding the cluster’s state as soon as new test data becomes available. Followed the process of continuous integration / continuous deployment (CI/CD).