AI Engineer

Simon Willison Portfolio

Creator of Datasette (11.5k GitHub stars) and the LLM command-line tool (12.6k), and a co-creator of Django per his bio. The blog has run since 2002 and posts most days on model releases, agentic coding and AI security incidents, with tag counts like python (1,284) and claude (313) shown right on the homepage.

LLM ToolingOpen SourceCLI ToolsTechnical Writing

Location

Didn't say. We're respecting the mystery.

Companies

Prime RadiantEventbriteLanyrd

Skills

LLM ToolingOpen SourceCLI ToolsTechnical Writing

Notable

  • Creator of Datasette, an open source multi-tool for exploring and publishing data (11.5k GitHub stars)
  • Creator of LLM, a CLI and Python library for language models (12.6k GitHub stars)
  • Co-creator of the Django web framework (per his bio)
  • Member of the Python Software Foundation board of directors (per his about page)

What makes it work

A breakdown of the choices that make this portfolio stand out.

1

A shipped, maintained CLI tool as the primary proof of skill

LLM installs via pip, Homebrew, pipx or uv, talks to OpenAI, Claude, Gemini and dozens of other models, and logs every prompt and response to SQLite. Version 0.36 shipped on Sept 22, 2026 with bug fixes from five new contributors. A maintained tool is a stronger engineering credential than a writeup describing one.

2

Fast, specific analysis of new model releases

Same-day posts like "Claude Opus 5.5, GPT-6 Sol, GPT-6 Luna, and a new price war" (Sept 22, 2026, written the day both launched) demonstrate hands-on testing, not secondhand summary — the kind of applied judgment that signals real usage, not just reading announcements.

3

An extensive tagging system across years of posts

Consistent tagging lets the archive function as a searchable knowledge base rather than a disposable feed, which is what makes years-old posts still get found and cited.

What AI Engineers can take from this

Specific, actionable tips to apply to your own portfolio — no generic advice.

  • Ship an actual tool, even a small CLI utility, rather than only writing about how you'd solve a problem.

  • When a new model or tool releases, publish your own hands-on test results quickly — firsthand testing is worth more than secondhand summary.

  • Tag and organize your writing consistently from the start so old posts stay discoverable as a reference, not just a stream.

Projects

Simon Willison presents his projects as release notes: the homepage mixes datasette 1.0a41 and llm 0.36 in with pelican benchmarks and a pixel-art kākāpō party. Nothing below is a showcase piece. Every one of them is still in the changelog this week.

Open SourceadvancedOngoing since 2017

Datasette

PythonSQLiteASGIJavaScriptWebAssembly (Datasette Lite)

An open source multi-tool for exploring and publishing data. Point it at a SQLite database and it serves an interactive, explorable website plus a JSON API. It is aimed at data journalists, museum curators, archivists, local governments and researchers. Datasette Lite packages the whole thing in WebAssembly so it runs entirely in the browser.

It shows the whole lifecycle of a real open source product: a plugin ecosystem, versioned docs, an alpha series toward 1.0, and outside contributors landing features. Maintaining that for years is harder than launching it.

  • Turns any SQLite file into a browsable site and JSON API with one command: datasette serve path/to/database.db

  • Plugin architecture: datasette-explain, datasette-auth-github and more are released as separate packages

  • Datasette Lite runs the full app in the browser via WebAssembly, with no Python server

Open SourceadvancedOngoing since 2023

LLM (command-line tool and Python library)

PythonSQLiteClickPlugin system

A CLI tool and Python library for working with OpenAI, Anthropic's Claude, Google's Gemini, Qwen, DeepSeek, Mistral and dozens of other models, through remote APIs or locally installed models. It runs prompts, stores every prompt and response in SQLite, generates embeddings, extracts structured content, and can grant models tools to execute.

It is AI-engineering infrastructure you can inspect: a provider-agnostic abstraction, a plugin interface for new models, and a local log database that makes prompts reproducible. It covers the design problems AI engineer roles actually hire for.

  • Installs via pip, Homebrew, pipx or uv; uvx can point it at any OpenAI-compatible endpoint without installing

  • Logs every prompt and response to SQLite, which Datasette can then browse

  • Model support lives in plugins (llm-anthropic and others), so new providers ship without core releases

Ready to build your portfolio?

Follow the AI Engineer roadmap — skills, projects, and timeline to get hired.

AI Engineer Roadmap