Rethinking software engineering levels

In mid 2019 I took on a 50%-time role as “Advancement and Learning Facilitator” in Broad’s Data Sciences Platform. One of my big projects in this role was to redesign our software engineering ladder in collaboration with our engineering managers.

➡ You can read the writeup of that process here.

A new file format for genomic data

Genomic data is huge - often hundreds of gigabytes per sequence file. For sequencing centers such as the Broad Insitute’s Genomics Platform, the cost of keeping historical data around can be very high, even when moving to cold storage.

I took a three month sabbatical to see if lossy compression could be used without affecting the data’s usefulness for scientific analysis. I wrote a compression algorithm that reduced file sizes to 4% and ~28% the size of its lossy competitor, and travelled to Basel, Switzerland to present it at the Global Alliance for Genomics & Health’s 6th Plenary Meeting.

Watch the talk or flip through the slides.

Reducing integration test runtime from 55 minutes to 15 minutes

Our integration tests relied heavily on creating clean Google Cloud Platform projects for test isolation. Project creation itself takes a few seconds but there were some group creation and population steps that were eventually consistent, emphasis on the eventually. This resulted in erratic test timeouts when Google was feeling slow that day, and slowly the timeout was increased as tests repeatedly failed.

One day I overheard someone say “I guess I’ll increase the timeout to 20 minutes”. I turned horror into initiative and wrote a tech doc to pool the projects within an hour. Within a week I’d built a prototype and within a month the microservice was deployed to our test environment. Our test runtime dropped over 70% as a result.

How much does Boston Police Department spend on surveillance equipment?

A project I did for the ACLU of Massachusetts’ tech division, cross-referencing BPD’s spending data (sourced from the City of Boston’s Checkbook Explorer) against companies listed on the Surveillance Industry Index.

GitHub repo with Jupyter notebook here.

Is there profit to be made trading SPAC warrants?

In late 2020 there was a craze in the financial markets for SPACs (special purpose acquisition companies, essentially shell companies that would search for a startup to merge with, launching the startup while avoiding the IPO process). Warrants are cheap financial instruments attached to SPACs, and are thus a way to potentially make big returns on small investments. There were many stories of traders doing this, so I scraped a lot of data to find out if trading them was generally profitable. The short answer is “yes, if you have a time machine” – soon after I did the analysis, the craze blew up, the market oversaturated, and this investment strategy got even more risky.

Writeup here.

A quick look at the 2017 Alabama special Senate election

The Doug Jones v. Roy Moore special Senate election in AL was close, and a particular breakdown of the vote totals went viral on social media. I decided to do a little digging of my own, and concluded that the biggest driver for the election was turnout, not race or gender.

Writeup here.

Other things

I also have many projects that got shelved, so if you’d ever like to discuss using FEC data to track political donations, or the time I wrote a Solidity parser while investigating the extremely scammy end of the crypto market, do ask!