Engineering Manager for the Homefeed Blending Team. The homefeed blending team is responsible for building out the best list of pins to show to the Pinner based on relevancy, diversity of the content and other business objectives. We use a variety of techniques such as Determinantal Point Process (DPP) to create a diverse set of pins for Pinner, Bayesian optimization for finding optimal tradeoff among different objectives. The team is currently working on applying a more ML learnt approach (reinforcement learning, list-wise optimization) to uplevel previous heuristic based rules. We also work closely with Ads team on whole page optimization and home product team on module blending to present a cohesive content experience to the user.

Led cross organizational effort to build out an offline experiment framework that simulates online ranking and compute experiment results offline. Significantly increased developer velocity by reducing time to collect data online from 2 weeks to offline in just a couple hours.Built out real time content distribution system using embedding based technology to solve cold-start problems for new content and new creators. Increased the amount of new creator content impressions by 80%. Drove 23% increase in gross merchandise value (GMV) of new shopping content.

Led and shipped various products on Android that are used by millions of users. Built out video upload capability and supported videos as a new content format. Increased video views by 10 times over 6 months. Iterated on Shopping and Explore tab features that led to 10% increase in user sessions.

Redesigned the Python-based Mapper to use Elixir supervised Python worker architecture. In this model, Elixir handles all of the network traffic and hands requests to Python workers for processing. Reduced the load on the ZooKeeper by 100 times, increased throughput by two times, saving more than $100k a year.Implemented distributed tracking, which involves creating log at each system checkpoint, writing a Spark job in Scala to upload the processed logs to HBase. Dramatically improved the visibility of the system by reducing the time required to track an event from writing complicated hive query to a click of button.Designed a high frequency rate limit service that serves thousands of requests per second. Used in production at Pinterest to protect downstream services from being overloaded.Led a cross team effort in reducing spam content in new user homefeed by 70%. Improved the availability of the spam checking service and added new heuristics in detecting spammers.

Model interactions of objects localized by RFID (Radio Frequency Identification). Developed machine learning algorithms to model human behavior pattnerns of interation with objects. Objects movement are tracked by RFID. The system alerts when the interaction pattern is abnormal. Identified features to improve classification accuracy and achieved an accuracy of 80%.

Explored options to store terabytes of advertisement data more effectively using column store. Resulted in data compression up to three times and map reduce queries sped up to eight times. Developed an application that analyzes advertising data in real time using Kinesis. Designed and implemented a visualziation that displays the real time data. First Place in Hackathon. Created an app that recommend recipes to users based on pictures that they took of raw ingredients.

Created a web app that presents Bridgewater’s insight on the market. Clients create their own portfolio via an interactive web UI. The risk and return of the portfolio are then presented to the client. Implemented the backend of the app in Matlab and the frontend in Javascript, providing a scalable and responsive design.Initiated and lead 4 other interns in organizing a company-wide, weekend-long hackathon.

Automated daily space data collection from major satellite company websitesAnalyzed large dataset for correlation between satellite anomalies and space weatherconditionsDeveloped an app for real-time satellite hazard quotient monitoring

Worked in team of 3 and created an app that analyzes acdemic writings of English asSecond Lanuage (ESL) researchers and make recommedations to improve their writingsParsed 1,000 HCI academic papers by native English spearkers from the past 5 years to bestored as a corpus for recommendationImplemented the algorithm that detects word usage errors in ESL writings