Curriculum Vitae

Overview

Ten years academic training and research work as a computational physicist and astronomer. Switched to the private sector in 2014 tackling machine learning and data problems at mid stage startups. Comfortable implementing prototypes in notebooks and writing production ready Python code. Strong understanding of data pipelining as well as model development, deployment, and evaluation. Experience at the individual contributor and team manager level. Passionate about continuous learning and open source development. Plays well with others.

Experience

Staff Machine Learning Scientist | Tempus AI | 2021-08 to Present

Applying machine learning to real world biomedical data for personalized medicine.

ML lead on the generative AI team focusing on oncology
Designing, building, and deploying models that reason over large collections of free text clinical notes
Robustifying systems in order to handle noise from messy real world data
Construction and cleaning of high quality biomedical knowledge graphs for Retrieval Augmented Generation (RAG)
Experimenting with multi-modal models combining structured electronic healthcare records (EHRs), clinical text, molecular sequencing, claims, and imaging

Board of Directors | HacDC | 2021-04 to Present

Helping to organize and run one of the oldest hacker / maker spaces in the US.

Served on the board in severeal roles including Vice President
Started the HacDC discord server and grew it to more than 500 users
Lead weekly machine learning by doing course and reading groups
Incubated and grew the hyperdemocracy project

Chair, Economics of Misinformation Working Group | Credibility Coalition | 2021-01 to 2021-10

Lead a group investigating the economic impact of misinformation in programmatic advetising.

Created the "AdSellers Dataset" composed of 40 million ads.txt records and 1.8 million sellers.json records
Published our findings in Examining Opaque Programmatic Markets with the Credibility Coalition AdSellers Dataset

Machine Learning Engineer (R&D Team) | Kensho | 2017-07 to 2021-07

Making geopolitical structured data and news streams useful in machine learning tasks.

Joined DC office as second ML engineer and eleventh employee as the office explored products aimed at geopolitical news analysts
Participated in product road mapping, hiring, customer pitches, rapid prototyping, and research on extracting structured information from text
Rapid prototyping for weakly supervised text classification. Experimented with zero-shot natural language inference / textual entailment and sentence transformers. Applied positive and unlabeled (PU) + confident learning (CL) techniques to data with noisy labels.
Researched, planned, and developed initial version of Kensho's Named Entity Recognition and Disambiguation (NERD) technology that maps mentions of named entities in text to knowledge base (KB) entries (mention extraction, co-referencing, candidate generation, feature generation, KB entry prediction)
Lead instructor for internal education group teaching transformer networks to Kensho employees.
Kensho lead on Wikimedia for Natural Language Processing (NLP). Developed a pipeline that synthesizes a consistent machine readable NLP dataset from raw Wikipedia and Wikidata dumps.
Applied Wikimedia dataset to R&D tasks such as synonym generation, word sense disambiguation, combining word and graph embeddings, custom topical corpus generation, and NER data augmentation.
Topological data analysis for text. Used the mapper algorithm (lens, cover, pull back, cluster) to produce compact graph-based representations of large corpora.
Designed semantic corpus search (e.g., search using keywords, concepts, entities, or entire documents).