I was responsible for the data platform at Zeenk, an e-commerce analytics company. I was responsible for the design/architecture and oversaw implementation of all of the backend systems. At this role I spent about 60% of my time as an individual contributor, and 40% of my time doing software design, code reviews, mentoring, and performing other management roles for my team of 4-6 platform engineers.Projects:- Designed and implemented a SQL and JSON API query service which automatically constructed and cached data cubes from large dimension and fact tables in a data warehouse, leading to consistant sub-second query performance for ad-hoc analytics queries against hundreds of thousands or millions of rows. This service powered the Zeenk BI product and could execute the queries as fast as the user could click on the UI to drill in/out. (Java, MySQL, Snowflake)- Designed and implemented a framework to easily write transform jobs to process raw marketing, sales, and user event data from different channels into a data warehouse with unified schema for analytics (Java, SQL, Spark dataframes, Snowflake)- Co-designed and implemented an orchestration system to execute data transformations in parallel on fixed schedules or in response to external events such as new raw data arriving (Python, Docker, AWS CloudFormation and CDK, AWS Batch, AWS Lambda, AWS EventBridge)- Designed and implemented a service to ingest large volumes of marketing and sales data from Google, Shopify, Facebook, TikTok, Amazon, on regular schedules and write to persistent storage (Java, AirByte Open Source, Amazon S3)- Designed and implemented an internal REST API service which served metadata to the rest of the backend services (Java, MySQL, Snowflake)- Co-designed and partially implemented a user graph system to flexibly track user events over time and assign attribution at query time (Python, SQL, DBT, Snowflake)

I designed and lead the implementation of most of the the software at Noumena. Our product was a suite machine learning tools for performing advanced causal inference, primarily for online advertising applications.Projects:- Designed and implemented the Timeline data structure for efficiently representing time series data for machine learning, writing a custom a generated code system to represent arbitrary data objects in memory as Java POJOs so they could be directly accessed using our feature engineering DSL.- Evolved Nanigans' ad-tech machine learning platform (which I wrote) into TQL (timeline query language), a general-purpose query service for extracting ML datasets from arbitrary heterogenous time series data.-Implemented Randall Lewis' causal modeling research on top of TQL, including implementation of many complex mathematical constructs in a distributed and scalable fashion. - Rewrote the Noumena Python library with an emphasis on easy understandability and usability- Made the entire Noumena product installable as a Python package (even though it uses Java under the hood). - Established architecture, coding style and documentation standards for the engineering team- Provided mentorship and a collaborative learning environment for others on the team to grow.

Nanigans was an online advertising company. After a short stint doing general purpose data engineering, I spent the majority of my tenure there building machine learning infrastructure for an RTB advertising system (user-by-user value predictions with high QPS and tight SLAs).Projects:- Designed and implemented a machine learning ETL system using Hive + Luigi, then evolved it to use Hadoop Map/Reduce, then Spark.- Contributed to the development of a DSL for feature engineering which gave the data scientist user a flexible, python-like syntax for declaring feature transforms, but compiled to native Java bytecode. - Contributed to the development of a real-time machine learning model evaluation service to score RTB bid requests that made thousands of predictions / second with <5ms response time. - Designed and implemented a REST API server for machine learning model metadata.- Wrote a distributed backend for an in-memory pivot table engine which became the stickiest and most used part of the Nanigans product.

I was one of two technical co-founders of Lumatic, which was started out of college with two friends. Initially worked on productizing computer-vision based geolocation, we went through the Techstars Boulder startup program in 2011, pivoted to building a turn-by-turn navigation smartphone app for walking and public transit directions, enhanced by annotated images of streets and landmarks. Lumatic was eventually acqu-hired by Nanigans in 2013.Notable projects I lead or contributed heavily to:- Backend API server development in Python/Django- Building and deploying a multi-modal routing engine including walking and public transit directions- Ingesting OpenStreetMap street data for use with our routing engine- Designing and implementing a spatially-sharded cloud computing architecture to serve navigation and POI requests.- Downloading, indexing, and storing 10M Streetview-like images - Downloading, cleaning, and storage of GTFS Schedules- De-duping, cleaning, and storing 15M entries of point-of-interest data (Yelp, 4Square, Facebook Places, etc)- Android app development

Worked as a lab research associate building laser spectrometers for applications in materials research

Worked as a lab associate building simulations of mitosis in human cells using C++.