profile picture

How to Query (Almost) Everything

Venue: "Have You Tried Rubbing a Database On It (HYTRADBOI) 2022" conference, remote, April 29, 2022

Presented the Trustfall engine's ability query arbitrary combinations of data sources: everything from APIs to databases to files on disk to ML models. One way to use this ubiquitous query capability is to prevent repeated debugging of the same problems by writing queries that look for bugs across the boundaries of different systems. For example: if using a Python codebase whose pyproject.toml configuration file specifies that Python 3.9 is required, but whose deployment information specifies a Docker image that installs Python 3.7. To prevent this problem, one can query for "Python projects whose required Python version does not match the Python version installed in their Dockerfile."

Here's what audience members thought of this talk:

Adopting Static Typing in Scientific Projects

Venue: Maintainers Track of the SciPy 2021 conference, remote, July 14-16, 2021.

Presented typing_copilot, a tool that simplifies the process of adopting static typing in a large existing Python codebase, together with a case study of using typing_copilot to begin type-hinting arviz, a library for exploratory analysis of Bayesian probabilistic models in Python.

Joint talk with Colin Carroll.

GraphQL as a High-Performance Cross-Database Query Language

Venue: Boston GraphQL meetup, September 10, 2019

Presented the GraphQL compiler project and the way Kensho uses it for cross-database querying workloads.

This talk was not recorded. However, the demo and slides from the meetup are available, and the presented work is also covered in more detail in three posts on the Kensho blog.

Compiled GraphQL as a Database Query Language

Venue: W3C Workshop on Web Standardization for Graph Data, in Berlin, Germany, March 4-6, 2019

I was invited to this workshop to present my work on Kensho's Knowledge Graph through the open-source GraphQL compiler project. The goal of the workshop was to gather graph querying practitioners across the technology spectrum for discussions toward the creation of a new query language standard ideally suited for cutting-edge query use cases.

Laika: Efficient In-Place Scheduling for 3D Mesh Graph Computations

Venue: 30th ACM Symposium on Parallelism in Algorithms and Architectures (SPAA), in Vienna, Austria, July 16-18, 2018

Presented a novel approach for performing complex graph computations such as finite-element analysis 7-13x faster than the prior state-of-the-art. This speedup is achieved through a near-optimal deterministic parallel scheduler that is better able to take advantage of increased CPU core counts, causes fewer cache misses, and has lower scheduling overhead than prior approaches.

Joint work with William Hasenplaugh, David Lugato, and James J. Thomas.