Paul Koch
Principal Research Software Design Engineer at Microsoft
Seattle, Washington, United States
Summary
🤩
Rockstar🎓
Top SchoolPaul Koch is a Principal Research Software Design Engineer at Microsoft based in Seattle with six years of focused experience in research-oriented software and machine learning tooling. He contributes to InterpretML, advancing practical interpretable ML, and works on a separate project optimizing transportation—bridging research algorithms with production-ready systems. On GitHub he enhanced the guidance project's LlamaCpp integration to enable local LLM usage, improving API compatibility and test-time memory performance. Trained as a computer engineer at the University of Waterloo, he blends low-level systems thinking with higher-level ML workflows. Paul’s work shows a knack for making cutting-edge model tooling both efficient and reproducible, especially in enabling local inference paths that reduce cloud dependency. Colleagues describe him as a pragmatic problem-solver who turns research prototypes into maintainable engineering.
6 years of coding experience
BrebeufBachelor's, Computer Engineering, Bachelor's, Computer Engineering at University of Waterloo
French, German
Stackoverflow
1reputation
0reached
0answers
0questions
Github Skills (9)
machine-learning10
lm10
large-language-models10
python10
llm10
pytest9
testing9
c-language8
cprogramming-language8
Programming languages (6)
Github contributions (5)
guidance-ai/guidance
Sep 2023 - Mar 2025
A guidance language for controlling large language models.
Role in this project:
Backend & ML Engineer
Backend & ML EngineerContributions:11 reviews, 48 PRs, 124 pushes in 1 year 5 months
Contributions summary:Paul implemented a `llama.cpp` interface to integrate large language models within the guidance framework. This involved creating a new Python module to interface with the `llama.cpp` library, enabling the use of local LLMs. Further contributions include updating and maintaining the `LlamaCpp` model to use the newer API for `llama.cpp`, and optimizing the testing environment by using a single cached model for the LlamaCpp tests. The user made adjustments to improve performance, particularly regarding memory usage during testing.
conda-forge/interpret-core-feedstock
Feb 2023 - Mar 2025
A conda-smithy repository for interpret-core.
Contributions:37 PRs, 24 pushes, 23 comments in 2 years 2 months
Find and Hire Top DevelopersWe’ve analyzed the programming source code of over 60 million software developers on GitHub and scored them by 50,000 skills. Sign-up on Prog,AI to search for software developers.