Hamel Husain Portfolio
ML engineer who worked at Airbnb and at GitHub (staff ML engineer, 2017–2022, and lead on the CodeSearchNet benchmark). His blog and teaching now focus on LLM evals, including the "AI Evals for Engineers and PMs" course he co-teaches with Shreya Shankar, which the site says has taught over 5,000 students from 500+ companies.
Location
Coordinates unknown. Vibes: international.
Companies
Skills
Notable
- First author, "CodeSearchNet Challenge: Evaluating the State of Semantic Code Search" (arXiv 1909.09436)
- Creator of fastpages, a Jupyter-notebook blogging platform under fast.ai (3.5k GitHub stars; archived, deprecated in favor of Quarto)
- Core contributor to Axolotl (LLM fine-tuning) and fast.ai's nbdev, fastcore and ghapi
- Co-teaches "AI Evals for Engineers and PMs" on Maven with Shreya Shankar
What makes it work
A breakdown of the choices that make this portfolio stand out.
A single, narrow specialty within a broad field
Rather than covering "AI" broadly, the site is almost entirely about one specific, high-value problem — evaluating LLM outputs — which is a harder and more differentiated position than generalist AI content.
Opinionated, contrarian post titles
"'It's Hard to Eval' Is a Product Smell" states a specific, debatable claim rather than a neutral how-to title, which is what gets a technical post actually discussed rather than skimmed.
A course sold with specific enrollment numbers
"Over 5,000 students from 500+ companies (including OpenAI, Anthropic, and Google)" is self-reported, but it is specific. That is a concrete measure of demand for the expertise being sold, not a vague claim of authority.
What AI Engineers can take from this
Specific, actionable tips to apply to your own portfolio — no generic advice.
Pick one narrow, hard problem within your field (not the whole field) to become known for — narrow and deep beats broad and shallow.
Write post titles that state a specific, debatable claim rather than a neutral how-to — it's what gets technical work actually discussed.
If you teach a course or workshop, publish real enrollment or outcome numbers rather than describing it only in qualitative terms.
Projects
Hamel Husain's open-source page is a table with a Role column (Creator, Core Contributor, Lead, Author), which says exactly how much of each project was his. Few portfolios are that precise about credit.
CodeSearchNet Challenge
A corpus, baseline models and benchmark for semantic code search: retrieving relevant code from a natural-language query. The corpus holds about 6 million functions across Go, Java, JavaScript, PHP, Python and Ruby, with 2 million (docstring, code) pairs. The evaluation set has 99 natural-language queries with about 4k expert relevance annotations.
Designing the evaluation (expert relevance labels, repo-level train/test splits to prevent leakage, a public leaderboard) is the same skill as LLM evals today, done years earlier.
About 6M functions across six languages, with 2M natural-language/code pairs
Splits data so code from one repository appears in only one partition, avoiding leakage
Community benchmark hosted on Weights & Biases, with logged runs and model artifacts
An easy-to-use blogging platform that turns Jupyter notebooks, Word documents and Markdown into a static blog through GitHub Actions. It lives under the fast.ai GitHub org and was also used to build the official Kubeflow blog.
It shows developer-tooling instincts: CI as the publishing engine, notebooks as source of truth. The clean deprecation shows the maturity to retire your own project when something better exists.
Publishes notebooks as blog posts automatically via GitHub Actions
Used to create the official Kubeflow blog
Deprecated in favor of Quarto, with a migration guide
More AI Engineer portfolios
Ready to build your portfolio?
Follow the AI Engineer roadmap — skills, projects, and timeline to get hired.