# Ryan Smith — smithai.me

> Personal site of Ryan Smith, a VP of Product who leads AI product teams in data-heavy platforms. He de-risks the hardest bets by prototyping the core himself, then builds and leads the team that takes it from prototype to production to org-wide adoption. U.S. Marine Corps veteran. Based in Georgetown, TX; works remotely.

This site is candidate-marketing material for VP Product and senior AI product leadership roles. It pairs written case studies with working, guard-railed AI features Ryan built himself. Metrics are framed conservatively; the AI assistant answers only from a vetted knowledge base and declines anything it can't source.

## Main pages

- [Home](https://smithai.me/): Overview — positioning, impact highlights, the work, approach, career timeline, and contact (Calendly + email).
- [AI playground](https://smithai.me/ai): Four working AI features Ryan built — a grounded assistant over his own work, a job-description fit reader, a live natural-language-to-SQL demo, and a human-in-the-loop book-metadata assistant that drafts commercial metadata from a chapter, excerpt, or draft blurb, with subject categories constrained to a verified subset of the real BISAC vocabulary. The page also publishes the features' unit economics (cost per answer, caching, hard spend ceilings) and their evaluation results: a golden-set eval graded by an LLM judge, with failures and known limits disclosed.

## Case studies

- [Agentic buying, from months to days](https://smithai.me/work/agentic-buying): An executive-sponsored initiative that compressed the media brief-to-activation cycle from months to days. Ryan prototyped the core agentic reasoning to de-risk it, then led the team that proved it in a multi-client pilot.
- [One AI assistant, adopted org-wide](https://smithai.me/work/ai-orchestrator): Unified fragmented AI efforts into a single multi-agent assistant — from personal prototype to executive buy-in to organization-wide adoption (3,500+ users).
- [Six hours to one, at global scale](https://smithai.me/work/billing-redesign): A reporting-and-billing backend redesign that cut the daily processing window from six hours to one at the same data volume and cut the data-error rate roughly 4× across every market.
- [Plain-language questions, trustworthy answers](https://smithai.me/work/conversational-analytics): A natural-language-to-SQL agent for non-technical users, designed with transparency and auditability so every answer can be verified.

## Writing

- [I built an AI tool that worked. I shut it down anyway.](https://smithai.me/writing/killing-my-own-product): The full story of the Jira tool Ryan killed nine months in — measured by fewer engineering complaints about ticket quality, shut down when PMs stopped verifying its output — and the verification principle now built into everything he ships.

## Resources

- [Résumé (PDF)](https://smithai.me/resume.pdf): Ryan's public résumé.
- [Résumé (Markdown)](https://smithai.me/resume.md): The same résumé as plain markdown — the easiest version for AI agents and assistants to read.
- MCP endpoint: https://smithai.me/api/mcp — a Model Context Protocol server (streamable HTTP, no auth) any MCP client can add as a custom connector. Tools: ask_about_ryan, check_fit (an honest job-description fit read that names real gaps), get_case_study, get_beyond_the_resume (technologies honestly tiered, a side build, values, working style — depth that doesn't fit a résumé). Same vetted corpus and honesty rules as the site's assistant.
- [LinkedIn](https://www.linkedin.com/in/smithrk81): Professional profile.
- Contact: ryan@smithai.me is the fastest way to reach him.