Vealth · Brandon Kelly
I am certified in construction safety. I also build the software.
Most people in safety do not write software. Most people who write software have never been accountable for a jobsite. I do both, and everything on this page can be checked by clicking it.
Los Angeles, California. Reachable at [email protected].
Certifications
Both issued by the Board of Certified Safety Professionals and accredited by ANAB to ISO/IEC 17024. Both current. Click either one to verify it at BCSP rather than taking my word for it.
Anthropic Academy certificates
Sixteen course certificates, completed March and April 2026. These are training completions, not accredited certifications, and I do not present them as equivalent to the two above. What they show is coverage: the full published curriculum, including the advanced Model Context Protocol, subagent, and agent skills material. Every one verifies publicly.
- Claude Code in Action
- Building with the Claude API
- Introduction to Model Context Protocol
- Model Context Protocol: Advanced Topics
- Introduction to subagents
- Introduction to agent skills
- Claude with Amazon Bedrock
- Claude on Google Cloud
- Introduction to Claude Cowork
- Claude 101
- AI Fluency: Framework & Foundations
- AI Capabilities and Limitations
- Teaching AI Fluency
- AI Fluency for educators
- AI Fluency for students
- AI Fluency for nonprofits
Built and running
Certificates say I studied something. These say I shipped it. All of them are live and none of them require my permission to inspect.
-
An ecological work protocol, onchainA contract on Base that posts bounded work, records claims, and settles completed packets. Deployed, source verified, and readable by anyone: 0x76c17C51336BE7B39F5164802e08b9811477A14B. Inspect it on BaseScan.
-
A free public timestamping registryAnyone can anchor any file's SHA-256 onto a public chain to prove they had it first. We take a hash, never a file, and verification is free by policy. Drop a file in and check it.
-
A Model Context Protocol serverA live MCP endpoint exposing the protocol's tools to any agent that speaks it. The server, and what an agent can do with it.
-
Pay-per-call HTTP paymentsEndpoints that answer HTTP 402 with a standard payment challenge and settle it without a middleman, so software can buy a single call without an account or an invoice. How the rails fit together.
-
A sealed safety record bookCrews file pre-task plans, daily reports, and corrective actions from a phone in about thirty seconds. A named human approves, never the AI, and each record is sealed the day it happens, so an auditor years later can trust the timestamp instead of the binder. Built and security tested. Not a compliance certification and not sold as one. See what a jobsite actually adopts.
What this combination is actually for
Safety software is mostly written by people who have never run a toolbox talk, and it shows.
I write web applications, agent systems, and verification infrastructure. The CHST is not decoration on that: it means a jobsite requirement arrives already understood, that I know what a pre-task plan has to survive, and that I can talk to a safety director and an engineer in the same meeting without translation loss. My organization is also approved under Anthropic's Cyber Verification Program for dual-use security research, which is an account authorization rather than a credential, and I mention it only because security review is part of what I do.
The work I am best suited to: safety and construction organizations that need real software instead of another spreadsheet, credentialing and verification platforms, and teams building agent systems who would rather hire someone who has already shipped one. Contract or employed, both fine.
If that is you, [email protected].
You do not have to hire me to use any of it.
Hiring is one door. The other is simply taking what is already built and running it yourself. These cost nothing and ask for no account:
- Anchor a file. Prove you had a document, a design, or a finding first, by putting its SHA-256 on a public chain. We take the hash, never the file, and verification is free by policy.
- Check somebody else's proof by dropping their file into the browser verifier. It never leaves your machine.
- Point an agent at the protocol over Model Context Protocol, and let it find work, read standards, and verify proofs on its own.
- Read how the rails fit together and rebuild any part of it. The method is written down, not hidden.
Built with Claude and GPT, and I will say so plainly
A one-person shop should not be able to ship a verified onchain contract, an MCP server, a payment rail, and a sealed record system. I can because two research labs did the hard part first. Anthropic's Claude does most of the daily engineering with me. OpenAI's models review the same code independently, which is how a solo builder gets a second pair of eyes that does not flatter the first.
I am not shy about that, and I do not think it lessens the work. Directing these models well is its own skill: knowing what to verify, what to refuse, and when the confident answer is wrong. Everything on this page was checked against the chain, the issuer, or a live page before it was published, because a model will happily tell you something is true. Credit to Anthropic and to OpenAI for the model development that made it possible. The judgment about what to ship is mine.
