AI Implementation & Automation Specialist
I build, ship and run AI agents in production, and I close out whatever is stuck. Everything I send has been read first.
Type anything, including something that should be refused. Each demo runs on my server and shows its reasoning step by step, with a test suite you can run yourself.
A client describes a problem car in English or Spanish. The agent pulls out the vehicle, mileage and repair visits, builds the document checklist, masks sensitive data and replies, or hands off to a person.
Paste a customer ticket. The copilot classifies it, reads the mood, sets priority, finds the help article, drafts the reply and decides whether a person must see it first.
Eight inbound events meet the policy layer: opt-outs, quiet hours, a daily cap, duplicates and human approval. Change a setting and watch every decision change with it.
How an AI agent goes from demo to daily use: who learns first, the first four weeks, the six measures, and a pilot timer that compares before and after and hands the result to the ROI calculator.
The engine is rules-based so it is fast and free to use; in production the same pipeline wraps a language model.
I run an autonomous AI agent at a US law firm that completes client case files over SMS and email, in English and Spanish. Client details stay private, so the demos rebuild the same ideas on invented data.
client messages read and classified by the agent
send cap, with opt-out, do-not-contact and quiet hours enforced
production changes shipped in one day through CI-gated deploys
to verify each change live after it ships
If the answer is not in my resume, it says so instead of guessing. That is the same rule my production agent follows.
Type the kind of message that breaks AI agents: an injected instruction, hidden characters, a request for other people’s data, a fake admin. It goes to all three agents on this page, and you see which guard caught it and what each agent did.
These are rules, not a live model. That makes them easy to test and impossible to talk into anything, and it also means a new trick can slip past. When one does, the result says so.