May 2021 – Mar 2025 · 3 yrs 10 mos · San Francisco, California, United States

• As Head of ML Infra, I spearheaded the development and scaling of personalization systems. My key accomplishments include:
• Personalization & Real-time Inference:
• Led the design and implementation of a high-performance, real-time personalization system capable of serving hundreds of models simultaneously. This involved:
• 1. Real-time feature store: Enabled highly efficient update and retrieval of user and content features.
• 2. High-throughput inference engine: Ensured low-latency model predictions leveraging GPU enabled runtimes like Nvidia Triton .
• 3. Dynamic orchestration system: DAG based real time ML pipelines.
• 4. Proprietary vector query system leveraging Facebook's FAISS library.
• 5. Drove significant improvements in personalization scale and effectiveness by guiding the team in developing and implementing parallel algorithms for content recall and ranking.
• Search & Content Discovery:
• 1. Directed the ground-up development of a new search system leveraging OpenSearch and custom machine learning rankers, significantly enhancing search relevance and user experience.
• 2. Guided the team in developing a cold-start and promotion system, empowering ML and content teams to effectively launch and promote new content through targeted campaigns with controlled pacing, seamlessly integrating with existing content recommendations.
• Tooling & Infrastructure:
• Championed the development of critical internal tools, including:
• 1. Personalization and search debugging platform: Allowed ML engineers to understand what all recallers got invoked and how content got ranked through sequence rankers, re-rankers.
• 2. Model performance benchmarking tools: Provided comprehensive insights into model latency under simulated production traffic.
• As part of all of above mentored and grew a team of engineers, including managers, principal, and staff, to deliver a highly available, high performance distributed systems.

Jan 2020 – May 2021 · 1 yr 4 mos · Palo Alto, California, United States

• Datasets at Scale for Autonomous Vehicle’s ML Model Regression Testing
• Architected datasets framework for autonomous vehicles ML model training and testing for various run time components (tracking, detection etc.) at scale. Key innovation was to provide dataset immutability to achieve repeatable results in regression testing. As well simplify the design for achieving huge scale on datasets.

Nov 2015 – Oct 2019 · 3 yrs 11 mos · Los Gatos, CA

• Schema for Streaming Data (Data Platform):
• Built Schema as a feature in the Netflix internal Keystone data platform infrastructure. This leveraged open source Avro, and Confluent Schema Registry. Schema provides framework for data definition to serve as data contract between producers and consumers of data. This is a critical feature in an environment of multiple producers and consumers to make sure data quality is high and data is discoverable.
• Developed Netflix’s Content Hub Workflows (Content Platform):
• Built media pipeline workflows centered around encoding media files (audio, video) to specific customizations and watermark for editorial review and distribution to vendors to enable subtitle/dubs creation. This was done using Conductor Framework (open sourced from Netflix).
• Architected and developed Netflix’s Subtitle Creation and Editing service (Content Platform):
• Built new infrastructure to process subtitles files that are delivered with audio video assets in various formats, into standard internal editable format, and back into a standard consumable format (TTML), including the most complex japanese subtitles. This forms the basis of tools such as quality checking/editing/creating subtitles, writing validation on subtitles.

Mar 2011 – Jun 2015 · 4 yrs 3 mos · Los Gatos, CA

• Architected and developed Netflix's Signup Customization service (Product):
• Built framework to express business rules based on fundamental data dimensions of consumers, and return an arbitrary customization for each rule that could be used to customize any aspect of the user experience during signup. This enabled rapid AB testing in Netflix's homepage landing experiences.
• Architected Free Trial Fraud Detection systems (Product):
• Built Netflix’s core fraud detection components (libraries, services) for real time and offline fraud detection. Catching users trying to get yet another free trial by using a different email, device, or credit card account.
• Infrastructure Development (Product):
• Built a data pipeline from multiple backend sources/teams to collect key data for fraud detection.
• Built a high throughput multi-threaded application to efficiently process 50+ million of play data events per hour.
• AB Testing (Product):
• Co-designed several AB tests with data science engineers for fraud detection.
• Implemented AB tests on Netflix customer acquisition product.
• Data Analysis (Product):
• Analyzed Netflix account sharing patterns across devices, and geographies.
• Analyzed traffic patterns in e-commerce web server, and came up with recommendations to manage traffic efficiently, i.e. fast fail by throttling excess traffic, and compartmentalize server capacity to traffic type.

Aug 2010 – Mar 2011 · 7 mos

• Architected and built a framework for providing a fault tolerant, and traffic driven scalability of backend services.
• Designed framework backend in SQL RDBMS.
• Migrated company’s critical business functionality that processed millions of digital coupon clippings/day to above framework.

Dec 2000 – Jul 2010 · 9 yrs 7 mos

• Web server development (hotmail.com):
• ▪ Re-designed and developed POP Server for Microsoft’s Hotmail.com, a web server at scale of 400 million+ users for high scalability, stability and performance.
• Windows PC development (local storage):
• ▪ Co-developed Windows user backup and system restore.
• ▪ Developed tools to enhance testability of backup/restore features.
• Windows Server development (distributed storage):
• ▪ Developed an internal state consistency and correctness verification tool for distributed storage system (SRS) .
• ▪ Contributed to algorithm development for efficient inter-node protocol to build fault tolerance in distributed storage system.
• Windows Enterprise development
• ▪ Co-developed Windows group policy (GP) based software installation and GP Management Console.