New to the EPP: Samson Musumuri
Samson Musumuri joined the Einstein Papers Project in the spring. He is filling a new role at the project, that of Data Scientist. We are grateful to have Samson on board with us. We asked him a few "getting to know you" questions. Enjoy!
Describe the work you were doing and/or the studies you've completed in recent years. How did those lead you to joining the EPP?
I did my master's in Analytics and Modeling at Valparaiso University in Indiana. After that I worked as a data scientist for about five years. Most recently I was a senior data scientist at Clearway Health near Chicago, and before that I was at Ventas in real estate analytics, at Cognizant, and at ESIC Hospitals in India.
Most of that work was the same underlying task. An organization has years of records in inconsistent formats, entered by many different people, and someone needs to make it usable. At Clearway, it was clinical text and pharmacy claims. Before you can build anything, you have to work out what the data actually means, and I like that part.
I was also teaching myself how large language models work, reading the papers, and building retrieval systems on my laptop to see what breaks. When I saw the EPP was hiring a data scientist, it was the same problem I already knew, in a much harder form. Over 100,000 documents, a lot of it handwritten German, a database holding decades of editorial decisions.
In what ways does your previous work dovetail with or diverge from what you are working on at the project?
The engineering carries over directly. Getting records out of an old database, reconciling fields filled in different ways by different people, finding near-duplicates, writing pipelines that give the same result every time. That is the same job whether the records are insurance claims or archival documents, and it is most of what I do here.
What changed is the tolerance for error. In industry, a system that is right about 95% of the time is a success, and the rest gets absorbed somewhere downstream. Here an error gets printed and cited. So, what I build does not produce answers. It narrows a set of candidates and hands them to an editor with a clear statement of how uncertain it is.
How have The Collected Papers of Albert Einstein (CPAE), or our archival database, piqued your interest while working at the EPP?
I expected the documents to be the interesting part. The editorial apparatus turned out to be just as interesting. Every annotation in a volume is a decision somebody made and could defend. That a letter was written in March, not April. That an unnamed colleague is a specific person. Those two fragments belong together. Most of that reasoning exists in the volumes, but not in any field a program can read.
The undated material interests me the most. There are close to two thousand loose pages of undated unified field theory calculations. Ordering them means using his scientific notation habits, the physical paper, and the mathematics. It is a hard problem, and one where computation and archival expertise each supply something the other cannot.
What interests and/or hobbies do you have? How did you discover or develop those interests?
Film is my main interest, particularly horror or science fiction. I track Formula 1 with attention, where the engineering is the story there, and the driver is one variable among many. I also play video games and read LLM papers during my free time.
What is a great piece of advice you have received?
One of my professors at university told me that if you go too deep into any subject, the result is zero, and that you have to know when to stop. I have found that to be true in almost every project I have worked on. There is always more you could check, more you could refine, another approach you could try. At some point the extra depth stops adding anything and starts costing you the work itself. Knowing where that point is has been more useful to me than any technical skill.