Ask about Steve. Get cited answers.
This agent answers questions about Steve Rolfe from a reviewed, public fact base, and every claim is cited to its source.
What has Steve built?
steverolfe.co runs seven live AI tools anyone can try: receipt-photo extraction, list cleanup and dedupe, document Q&A with cited sources, meeting-transcript notes, email diagnosis, email migration planning, and reconciling two lists against each other. A remote MCP server lets other AI agents call them as tools. The site and its first tool went from concept to live production in one evening. Steve also founded and built T3AM (t3am.app), a youth-sports team platform that went live about 6.5 weeks after its first commit. Measured from the repository on 2026-09-09, it has 430 merged pull requests, 1,450+ passing tests, and 19 lint-enforced domain modules, and it is WCAG 2.1 AA audited. The demo is at t3am.app/demo, with no account needed and access code 13579.
How does he handle AI hallucination?
His systems flag uncertainty instead of guessing. Low-confidence cells are marked for review, ambiguous merges stay unmerged, and meeting action items never get invented owners. Document answers use citation grounding, so each citation is a verbatim quote extracted by the API. Model output is checked against a schema before a user sees it. His working rule is stage, review, then commit, and he has documented cases of models misstating facts about his own products.
What's his experience with data pipelines?
He automated daily reconciliation for a registered investment adviser against its Orion system of record, reconciling roughly 350 accounts daily. It surfaces only the exceptions a person needs to touch, and the operations team checks it every morning. A companion pipeline pulls daily custodial transaction files over SFTP. It applies nothing when a batch has errors, excludes duplicates, can be reversed, and writes a per-run audit trail. Both run in production as containerized scheduled jobs on Azure. He has also built ETL between line-of-business systems, platform-to-platform migrations, and a FactSet-fed research platform that ranks the S&P 500 and Russell 1000 nightly.
Has he used AI in a regulated firm?
Yes. In 2026 he introduced AI-assisted development at an SEC-regulated investment adviser and directed Claude Code to build six production systems: a daily reconciliation, a custodial import pipeline, a research platform for the investment team, a pre-market brief for the portfolio manager, an operations dashboard with a natural-language Ask panel, and an allocation drift check. They run as containerized scheduled jobs on Azure, with the dashboards behind Microsoft SSO. A data classification set before any feature was built decides what may enter a prompt. Five of the six were his own initiative rather than assigned work. The pre-market brief is the exception, and he built that one too.
Why trust a solo engineer with production systems?
He has run the entire technology function of an SEC-regulated investment adviser for thirteen years. At its peak the firm had as much as $12 billion in assets under advisement across 7,000+ client accounts, and security, compliance, and disaster recovery fall to whoever owns the systems. When Morgan Stanley ran client due diligence, the reviewer skipped his prepared questions because the documentation package had already answered them. Earlier in his career, as Change Management Manager at a web platform that shipped updates daily, every update from seven or eight developers ran through his approval. On T3AM, every change goes through a pull request with CI gating the merge, and 1,450+ tests pass.
How do I contact Steve?
Email hello@steverolfe.co. He is based in St. Albans, Missouri, in the St. Louis area.
Every question runs through three separate AI stages. A boundary agent classifies it first and routes anything the agent should not answer to Steve. A grounded answerer writes the reply using Anthropic citation grounding, so every citation is a verbatim quote from the reviewed fact base. A critic agentthen checks the draft against the sources and Steve's accuracy rules before you see it, and repairs or blocks anything unsupported.
The fact base was reviewed and approved by Steve before publication. Questions and answers are processed per request and not stored. To talk with Steve directly, email hello@steverolfe.co.