Sep 2021 – Dec 2022 · 1 yr 3 mos

• I run Product & Customer Success for Tecton
• Tecton provides ML Platform with infrastructure to enable data teams to quickly and reliably build real-time ML use cases leveraging previously hard-to-use real-time and streaming data, while ensuring performance and optimizing cloud costs.

Oct 2020 – Aug 2021 · 10 mos

Dec 2019 – Sep 2020 · 9 mos

Apr 2015 – Dec 2019 · 4 yrs 8 mos

• Split sold for 9 figures to Harness. Not too bad!
• Split is a unified solution for feature flagging & experimentation

Feb 2013 – Apr 2015 · 2 yrs 2 mos · San Francisco Bay Area

• RelateIQ was a phenomenal learning opportunity. I built RelateIQ's data infrastructure including standardization of object models, automatic tracking, experimentation platform, hadoop infrastructure, and reporting. Building this infrastructure from ground up taught me valuable lessons in system design, the art of convincing others and building consensus for infrastructure improvements.

Nov 2012 – Feb 2013 · 3 mos

• My role has expanded to design and execute on a wider array of data driven projects within LinkedIn's 'Search, Network, and Analytics' team.
• I am currently focused on the following major priorities for LinkedIn:
• Improving the search experience for recruiters by increasing the speed and scalability of backend search systems.
• Rethink our content processing pipeline to allow a more distributed model of computation that reduces feature development time.
• Improve LinkedIn's ability to change user experience (in search, recommendations, and beyond) based on realtime user actions.

Apr 2011 – Nov 2012 · 1 yr 7 mos

• Working in ads, we realized the need for reacting in realtime to the actions of a user. If a user has just viewed an article about 'Mercedes Benz', maybe it's a good time to show her an ad for Mercedes Benz. The same principle could be applied to other domains within LinkedIn. For instance, if LinkedIn recommended a job to you that you haven't clicked upon, why not discount the ranking of that job. There was no system at LinkedIn that could help us answer this question in near-realtime. Therefore, I built one: a scalable near-realtime OLAP system to answer segmented count queries.
• For the first half of my stint as a Staff Software Engineer, I led the engineering behind our click prediction engine for direct ads. This was by far the toughest project I had worked on since this was the first project I built from the ground up in my stint at LinkedIn. The systems aspect was tough, the prediction aspect even tougher. Click through rates on advertisements are generally rare and predicting them is notoriously difficult. I did not lead the data science behind this modeling; however, I actively participated in feature engineering (ads in your social network) and devising model evaluation techniques.

Dec 2009 – Apr 2011 · 1 yr 4 mos

• There is a lot of value in the data people provide to social networks like LinkedIn. In my current role, I work on unlocking this value and providing useful insights to LinkedIn users by building a real-time recommendation engine. This engine powers the following products on LinkedIn:
• Jobs You May Be Interested In (recommend jobs to people)
• Groups You May Like (recommend groups to people)
• Companies You May Want To Follow (recommend companies to people)
• Talent Match (recommend people to jobs)
• Similar Jobs (recommend jobs to jobs)
• Similar Groups (recommend groups to groups)
• ... and of course you can think of other variants we might be working on but can't disclose at this point.
• The way my team works, we have people building the horizontal system (the engine itself) to power all these products. Then we have folks who focus on individual verticals atop this horizontal structure. My role has evolved into a horizontal platform oriented one where I am responsible for building the system that powers all these recommendations. Lately, though, I have also been focusing on owning a vertical product (matching people and ads) which is very challenging.
• What have I learned in this position? How to actually build a scalable and fast system. My positions at Yahoo! were more analytically oriented. I knew how to extract insight from data, but my current role has enabled me to pair that ability to extract insights with the ability to serve those insights through a fast, near real-time, scalable system.

Mar 2009 – Nov 2009 · 8 mos

• Engineer for Traffic Protection ("TP") team that fights click fraud on paid search traffic for Yahoo!. I wrote the TP performance monitoring system using Apache Pig over Hadoop processing terabytes of traffic data. The new system reduced reporting times from a week to less than a day. I also used Apache Mahout - a Map-Reduce implementation of Machine Learning algorithms - to cluster clicks to power TP machine learning models. I extended Mahout to measure cluster validity and pick ideal 'k' values for k-means. Lastly, I prototyped detection of botnet attacks and publisher coalition attacks; one of the most difficult click frauds to detect.

Aug 2007 – Mar 2009 · 1 yr 7 mos

• Lead Engineer for Keyword Research Service (“KRS”) team that uses collaborative filtering to suggest biddable keywords to Yahoo! Advertisers. I was responsible for improving relevance and coverage of keyword suggestions, reduce noise in keywords extracted from advertiser landing pages and rewriting the entire KRS datapipeline in Apache Pig.

Jan 2007 – Aug 2007 · 7 mos

• Aided research on rule making by implementing categorization of individual sentences withing larger documents. Used Support Vector Machines on a rich feature set.

Aug 2006 – Dec 2006 · 4 mos

• Created a framework for focused crawling on archived web pages using Heritrix, a Java web crawler

Mar 2005 – Jul 2006 · 1 yr 4 mos

• Too far back in the past to be relevant anymore.