Julie Tibshirani

Hello, world! I’m a software engineer focused on search and data-intensive systems. I currently work on code search and AI-assisted software development at Sourcegraph, and serve as an Apache Lucene committer and PMC member. I’m also the author of the generalized random forests (grf) package.

I love building software that’s empowering and joyful to use. I’m an open source enthusiast, and shared my experience joining the open source search community in Finding a home (and career) in the open source community.

Before Sourcegraph, I worked at Elastic on the Elasticsearch search engine, and at Palantir Technologies. I hold an M.S. in Computer Science and B.S. in Math from Stanford University.

Selected work

🔎 Vector search

Thanks to a new generation of machine learning models that can powerfully represent text as vectors, there’s been a surge of interest in vector-based semantic search. I led Elastic’s effort to introduce vector search in Lucene and Elasticsearch, helping extend these systems to become powerful “vector databases”.

• Introducing approximate nearest neighbor search in Elasticsearch
• Implementing academic papers: lessons learned from Elasticsearch and Lucene
• Text similarity search with vectors in Elasticsearch

🌲 Machine learning for causal inference

Causal inference allows for determining the effect of an action on a larger system. The generalized random forests (grf) method combines insights from statistics and machine learning to enable causal analysis. I authored the grf software package, which won an inaugural Stanford Open Source Software prize for its research impact, quality, and dedication to open source principles.

• Generalized Random Forests (Annals of Statistics, 2019)
• Local Linear Forests (Journal of Computational and Graphical Statistics, 2020)

💾 Information extraction

During my Master’s degree I researched relation extraction and knowledge base population as part of the Stanford Natural Language Processing group. Our work focused on non-traditional supervision techniques, including multi-label learning, partial supervision, as well as methods to address labelling errors in training data.

• Multi-instance Multi-label Learning for Relation Extraction (EMNLP, 2012)
• Combining Distant and Partial Supervision for Relation Extraction (EMNLP, 2014)
• Robust Logistic Regression using Shift Parameters (ACL, 2014)

Personal

⛰️ Backcountry cooking recipes. I’m an avid backpacker and enjoy finding creative ways to eat well outdoors.

📼 Digital mixtapes. A true child of the 90s, I love making playlists for friends and family.

• A Golden State: A rock-centric playlist featuring artists from California, compiled as a welcome gift for a friend moving here.
• Sleepytime: My daughter’s bedtime playlist, inspired by three prayers I said before bed as a child.

🤓 Unicode. I’m the proud sponsor of Unicode characters μ and σ.

---