--- title: Lucien George | Developer and agent resources description: "The machine-readable surface of luciengeorge.com: llms.txt, llms-full.txt, a markdown twin for every page, Accept: text/markdown negotiation, JSON-LD, and the open-source repository behind the site." url: https://www.luciengeorge.com/developers --- # Lucien George | Developer and agent resources This site is built to be read by software as well as by people. Everything below is a stable, public URL, and all of it is the same content the pages show. ## Machine-readable index - [/llms.txt](https://www.luciengeorge.com/llms.txt): the link index. Start here to see what exists. - [/llms-full.txt](https://www.luciengeorge.com/llms-full.txt): every section's markdown concatenated into one file. The cheapest way to read the whole site in a single fetch. - [/index.md](https://www.luciengeorge.com/index.md): the entry point for the site root, with the page map and guidance on when this site is the right source. - [/agents.md](https://www.luciengeorge.com/agents.md): instructions for an agent, including when to use this site and when to look elsewhere. - [/sitemap.xml](https://www.luciengeorge.com/sitemap.xml): every canonical URL with a `lastmod`. - [/robots.txt](https://www.luciengeorge.com/robots.txt): the crawl policy. AI crawlers and user-triggered agents are named and allowed. - JSON-LD (schema.org `Person`, `Organization`, `WebSite`, `FAQPage`) is inlined in the `` of every page. ## Markdown for every page Every page has a markdown twin at the same path plus `.md`. For example [/work/fyxer](https://www.luciengeorge.com/work/fyxer) is also [/work/fyxer.md](https://www.luciengeorge.com/work/fyxer.md). The canonical URLs also negotiate markdown, per [acceptmarkdown.com](https://acceptmarkdown.com) and RFC 9110: ask for `text/markdown` and you get markdown back from the same URL you found in search, with `Vary: Accept` so a cache keeps the two representations apart. Browsers still get HTML, because the header is ranked properly (q-value, then specificity) rather than substring-matched. Every markdown response carries YAML frontmatter with `title`, `description`, and the canonical `url`, so a fetched file can be attributed back to its page. A wrong guess at a URL returns a real 404 whose body names where to look next, rather than an app shell with a 200 on it. ## Source code The site is open source: [github.com/luciengeorge/lucien](https://github.com/luciengeorge/lucien). It is a TanStack Start app on Vercel with Convex behind it, and the assistant answers out of a retrieval index built from the same markdown that renders these pages. The repository carries `AGENTS.md` and `CODEBASE_ARCHITECTURE.md` at its root, which is what a coding agent should read before changing anything in it. The article [A portfolio that answers questions about me, gated by an LLM judge](https://www.luciengeorge.com/writing/rag-portfolio-with-a-blocking-eval-gate) explains how the retrieval and the eval gate work, with the parts that turned out to be wrong left in. ## The whole surface The files above are all of it. This is one person's portfolio rather than a product, so there is nothing to authenticate against, nothing to install, and no account to create. If you want to know something about Lucien George, the markdown is the interface, and it is meant to be enough. ## Attribution Quoting or summarising any of this is fine. Cite the canonical page URL rather than the `.md` twin, so a human following the citation lands on something readable.