Hamel Husain
ML engineer and consultant at Parlance Labs; earlier applied ML at GitHub and first author of CodeSearchNet
- ML engineer and independent consultant, Parlance Labs1
Hamel Husain is a machine learning engineer and independent consultant at Parlance Labs who helps teams debug, analyze, and measure their AI systems, work Husain calls evals, and co-teaches a course on AI evals for engineers and product managers. Husain previously did applied machine learning at GitHub and Airbnb; Husain's homepage links early LLM research from that time, the CodeSearchNet Challenge paper (2019), of which Husain is first author. At GitHub, Husain also wrote the 2020 GitHub blog post introducing nbdev, a literate programming environment for Python that GitHub helped fast.ai develop. Husain is a co-author, with Eugene Yan, Shreya Shankar, and others, of "What We've Learned From A Year of Building with LLMs" (2024).12346
Work
CodeSearchNet Challenge (external site: arxiv.org)Open data
First author of "CodeSearchNet Challenge: Evaluating the State of Semantic Code Search" (2019), which released the CodeSearchNet Corpus of functions from open-source code in Go, Java, JavaScript, PHP, Python, and Ruby, with expert relevance annotations for evaluating code search.32
nbdev (external site: github.com)Open-source code
Wrote GitHub's 2020 blog post introducing nbdev, a literate programming environment built on Jupyter notebooks that GitHub assisted fast.ai in developing, with documentation generation and continuous integration through GitHub Actions. The repository is licensed under Apache 2.0.45
What We've Learned From A Year of Building with LLMs (external site: applied-llms.org)Published research
Co-author of the 2024 essay on lessons from building LLM products, published on applied-llms.org and by O'Reilly in three parts.6
Other documented roles
Roles the sources document but that are not confirmed as current.
This profile covers public professional work only, from the sources below. It is not an endorsement or a ranking. To correct or remove it, use the correction form.