Skip to content
Holger Teichgraeber
Holger Teichgraeber

Holger Teichgraeber

Software Engineer
Door Dash
Bio

I am a Data Science Manager at Archer. Here, I work on the machine learning models, optimization models, and the software suite that support planning and operations, including Prime Radiant (see here for more details). Prime Radiant is Archer’s data science technology that allows us to better understand how people travel within cities around the world and informs the company’s strategic decisions on vehicle design and go-to-market strategy.

Previously, I worked as an Operations Research Scientist in the Data Science team at Convoy. There, I worked on building efficiency into our network of shipments. I developed, deployed, and maintained machine learning and optimization algorithms in production, and I provided quantitative insights to inform product decisions. A high-level description of the problem my team is solving for can be found here.I received my Ph.D. from the Department of Energy Resources Engineering at Stanford University. My advisor was Prof. Adam Brandt and I was a Wells Family Stanford Graduate Fellow.

In my Ph.D. research, I focused on applying state-of-the-art computational tools at the intersection of machine learning and optimization to energy systems problems. As an example, I have worked extensively on the development of new algorithms and applications of time-series aggregation for infrastructure planning and operations. Out of my research, two open-source software packages have emerged: TimeSeriesClustering implements unsupervised learning methods for time-series data, and CapacityExpansion provides an extensible, data-driven infrastructure planning tool for energy systems.


Last-mile Delivery Optimization in Production - An Experiment-based Framework

The assignment problem in last-mile delivery logistics platforms enables the connection of consumers, drivers, and merchants in the most efficient manner possible. The problem is solved in near-real-time and is prone to a high level of uncertainty. In this talk, we explore how to design and evolve this optimization algorithm in production. We discuss an iterative, experimentation-based framework that allows for fast iteration cycles, and highlight the framework’s advantages over the traditional approach of designing optimization algorithms through offline prototyping. We also discuss what skills in addition to traditional optimization modeling knowledge will make for a successful Operations Research Scientist in such real-time production environments.

Essential / Professional