Data Integrity - Led a team to architect and develop- from scratch - a comprehensive suite of data-integrity solutions for end-to-end corruption protection - across our multi-binary - distributed stack. This included multiple new infra-level products developed in conjunction with sister teams - as well as their inclusion into Firestore, with

Protobuf-checksumming - Designed, delivered and supported a solution to the thorny problem of protecting protobufs - as a part of the above effort. This is now not only deployed throughout Firestore, has also seen adoption across multiple products inside Google. It is currently the best solution for cross language, cross binary, robust protobuf checksumming.

Driving Safety - making Uber safer via AI and ML - for riders, drivers and everyone on the Uber platform!

We do a lot of Telematics ML Research at scale - using device telematics data at planet-scale to pinpoint possible safety-related incidents and inferences to possibly predictions and someday prevent them.

Tools I use: scala, python, spark, flink, golang

HARC: Designed and built a failure-resistant, distributed datastore solution around Redis to be consistent, elastically scalable, low-latency, and resilient to machine failure - to store and serve RCA bidding data

RCA: Worked on a fast, robust, scalable distributed pipeline to stream, process, aggregate and serve all RTB bidding data from our edge servers to internal customers cutting our micro-second latencies to milliseconds. Using Kinesis, Scala, Play, DynamoDB, Redis, AWS ELBs, ASGs, Terraform

On the side, I worked with our data-anthropology team to work up some fascinating analysis of browsing data :)

Velocity Fraud model: Built a velocity model that tracks growing clusters of suspicious customer behavior and flags ongoing large-scale fraud attacks as they ramp up

Mix-Media Marketing Attribution: Using our data on our media spend to attribute customer acquisition over the TV, email, mail, display advertising, and using time variable windowed features to attribute effectiveness and delay of ad-spend

Geo-targeted Campaign Lift Model: Developed a model to measure the uplift from a marketing campaign targeted at specific geographic area