I build the AI systems enterprises actually run on — autonomous agents, private RAG, and custom models that ship to production.
Multi-step agentic systems that run end-to-end workflows reliably — not demos that die in staging.
Fine-tuned models and custom RAG, deployed securely inside your own private infrastructure.
The pipelines and architecture that make your proprietary data clean, connected, and AI-ready.
fintech
representative engagement
healthcare
representative engagement
logistics
representative engagement
A payments team was burning 40+ analyst hours a week manually clearing back-office exceptions — stale vendor records, amount mismatches, duplicate holds. Volume doubled; headcount couldn't.
An autonomous agent with read/write access to the ERP through guarded tools. It diagnoses each exception, gathers evidence, resolves what it can prove, and escalates the rest with a full trace attached. Every action is auditable.
Clinical staff were answering policy and protocol questions from a 9,000-document intranet — or worse, from memory. Search returned documents, not answers.
A private RAG assistant deployed entirely inside the hospital's infrastructure. Answers cite their sources inline; anything below a confidence threshold refuses and routes to a human. Zero PHI leaves the network.
A logistics operator missed thousands of scheduling calls a month outside office hours. Every missed confirmation became a failed delivery attempt — real money.
A voice agent that answers, books, reschedules, and confirms appointments end-to-end, 24/7. It reads and writes the dispatch calendar live, sends SMS confirmations, and hands complex disputes to humans with full context.
Not screenshots — working systems. Press play.
"We had something running on our own data in three weeks. Every vendor before him had sent a deck and a six-month timeline."
"He was the first person to show us how the system fails before telling us how it works. That's the reason it's still in production a year later."
"He didn't hand us a system and disappear. He stayed on the hook for it until our own team could run it without him."
Two weeks inside your workflows and data. We leave with a map of where AI actually pays — and where it doesn't. Deliverable: scoped build plan.
A working slice on your real data in 2–4 weeks, not a deck. You interact with it, your team breaks it, we learn. Deliverable: live prototype.
Before anything ships: eval suites, guardrails, failure-mode testing, cost modeling. This is the step everyone else skips. Deliverable: evidence it works.
Deployed in your infrastructure with monitoring, tracing, and a runbook. Then I stay on the hook for it. Deliverable: a system you own.
Two weeks inside your workflows and data. You leave with a scoped build plan and an honest map of where AI pays — and where it doesn't.
BOOK THE AUDITPrototype to production on your infrastructure: architecture, evals, guardrails, monitoring, runbook. Scope sets the number.
SCOPE A BUILDOngoing tuning, evals, and monitoring once the system is live. Models drift and data changes — someone has to stay on the hook.
TALK RETAINERYour team leaves the day with something working on their own data. Not a workshop deck.
I'm a CTO who ships. For fifteen years I've built the systems companies actually run on — and for the last four, the AI layer on top of them: custom models, autonomous agents, and the data infrastructure that makes them trustworthy.
Forged In AI is deliberately small. You work with me — the person who architects, builds, evaluates, and stands behind the system in production.