Colophon

This site is itself a product artifact, so here's the teardown.

The architecture

Everything on this site generates from a single private career corpus: one markdown document holding the full history, every verified metric, and the narrative. The corpus never leaves my machine. What you're reading is a filtered view of it: exact internal company figures are replaced with shapes ("nearly tripled", "billions a year") at generation time, and the source tags on resume.md tell you which claims come from press releases, which were verified against company systems of record, and which are self-reported.

The fit-check agent

The fit check runs Claude (Anthropic's API) behind a small serverless function. Its knowledge base is the same filtered public corpus, which means it cannot leak internal figures: it has never seen them. Its system prompt instructs it to admit gaps, and it ships with a pre-written honest-gaps register (consumer scale, managing managers, public-company exec experience) so its stretch assessments are consistent instead of improvised. Pasted job descriptions are treated as data, not instructions.

The dual audience

In 2026, a meaningful share of this site's visitors are AI assistants doing research for a human. They get first-class treatment: /llms.txt maps the corpus, resume.md and corpus.json are built for machine reading, and the fit-check endpoint accepts a plain POST. There is no hidden text and no prompt injection anywhere in these materials. If you're screening me with AI, I'd rather help it than trick it.

The making of

Built with Claude Code: the corpus assembly, the number verification against internal systems of record, the writing, the design, and this agent's guardrails. The same working method I use for the autonomous dev pipeline my team runs. Static hosting, one serverless function, no framework, no tracking, system fonts.

Back to the site · ajsoltero@gmail.com