Judd Keppel

hacker, founder.

machine learning, javascript, decentralization.

for hire ($120 / hr)

github

twitter

about me

I’m a founder / engineer in Seattle. In 2016, my startup (Send) was acquired. Now I’m usually working on machine learning projects that I find interesting, such as:

My time is also for sale (currently $120 / hr). DM me on twitter! I can build your web / mobile app, train models on your datas, or help you sell your thing to your customers.

past projects

future projects

Stuff I find incredibly interesting and hope to explore soon in some projects:











~









project: send, inc

At Send, Justin Gough and I built software to help pharmacies & hospitals manage the delivery of mission-critical stuff, like:

By the end of 2015, we’d exceeded our goal of $1m in revenue for the year. We sold the company to Alpha Transportation in February 2016.

Our customers primarily interacted with us via desktop web on their work computers. The web frontend was built with react and browserify.

We also built an app for the delivery drivers in our network. Since the drivers used both iOS and Android, and it was critical to keep their apps working correctly, it seemed obvious and pragmatic to build an app where most of the functionality was contained inside a WebView (and thus worked the same on each platform, albeit with a slightly diminished UX). The Send driver app was built with Ionic.

As we scaled up and added employees to our ops team, we found ourselves building more internal tooling. We ended up building most of the things you might imagine a big O2O startup (Uber) would build for themselves. As we built tools, we’d add them to a centralized internal dashboard.

We also built a native iOS app for our customers with react-native. We experimented with providing services outside medical delivery to attract consumers who might use our service personally (not profitable), then advocate for us within their enterprise (very profitable).

The most interesting part of our software, though, was the routing backend. If you’re into computer science, the formal problem we needed to solve was a Vehicle Routing Problem (specifically, the Vehicle Routing Problem with Pickup and Delivery and Time Windows).

Our solution involved a manually-tuned cost function (weighted sum of pickup delay, delivery delay, and expected lateness) and genetic search for route candidates.

When we were acquired, I was looking at integrating predicted demand (as a function of price) into our VRP solver. The VRP solver server was written in node.js.