AI Systems Architect & Consultant
AI systems that run real businesses.
I design, build, and operate applied AI for small business operations — custom EMR work, CRM automation, AI voice and chat agents, KPI command centers, and autonomous daily agents. Everything described below is deployed and in production, not a proposal.
Book an intro callWhat I Build
Four systems, one architecture.
Custom EMR & Clinical AI
Custom electronic medical record layers on Canvas Medical: an ambient AI scribe that transcribes locally and drafts notes for clinician review, insurance card OCR with staff verification gates, a branded patient companion app, and controlled-medication inventory tracking inside the EMR.
Business Command Centers
One hub for the whole business: insurance billing tracked from submission through paid, pending, and blocked; cash-pay revenue current and forecasted; daily and weekly marketing reports on leads, conversion, and cost per lead; and captured SOPs staff can query like a veteran employee.
AI Agents
Inbound and outbound AI phone agents that answer, qualify, and book; website and SMS assistants that work from your full knowledge base with structured staff handoff; and daily reporting on every AI call and message — attribution, conversion, and booking outcomes, delivered every morning.
Automation & Reconciliation
Custom MCP connectors that give AI direct, governed access to your systems; scheduled daily agents for KPI tracking, cancellation and retention analysis, and inventory follow-ups; and reconciliation workflows that audit billing against the record and surface claims that were never submitted.
Case Study
Allay Health & Wellness + NeuPath
Over the past year I served as AI Systems Architect for a multi-site interventional psychiatry group in Palm Beach Gardens, Florida — building the technology stack that operates two clinics end to end.
A custom EMR, built from scratch on Canvas Medical
I led the clinics' migration off a closed legacy EHR and architected a fully custom medical record layer, live today with the full patient panel. On top of it: an ambient AI scribe that transcribes on the clinic's own hardware so audio never touches the cloud, a bidirectional CRM-to-EMR bridge that converts qualified leads into patient charts and unifies messaging into a single thread, a branded patient app with mood tracking, journaling, and secure messaging, and a Spravato inventory tracker managing the full lifecycle of a REMS-regulated medication — lots, expirations, orders, and schedule-aware reordering.
The Allay Command Center
The single place leadership sees how the business is actually doing: insurance claims from submission through paid, pending, and blocked; cash-pay revenue current and forecasted; automated marketing reports across Google, Facebook, and SEO so spend decisions run on this week's data; and an agent runtime where morning leadership briefings, alerting, and call classification run on schedule without a human pressing a button.
AI voice, SMS, and web agents across both clinics
An AI communications layer deployed across Allay and its sister clinic NeuPath: phone agents that answer, qualify, book discovery calls, and work lead lists on compliant calling windows; a website and SMS assistant with crisis-language escalation built in; and a custom CRM connector producing daily attribution and performance reports on every AI call and message. I QA the agents against real transcripts, not vendor summaries, and hold vendors to the numbers.
Daily agents on the medical record
A custom, read-only AI connector to the EMR runs scheduled daily agents: appointment and KPI tracking, cancellation and retention analysis, and inventory management that generates the day's follow-up tasks. An AI reconciliation workflow audits the billing team's monthly reports against the clinical record and surfaces claims that were never submitted. That is found revenue, every month.
Proof of Practice
The same architecture, applied to my own life.
I hold my personal systems to the same standard I sell. All of my personal agents run through Jarvis, a private, local-first AI hub I built that orchestrates my custom connectors and daily agents in one place.
Finance
A custom Plaid connector — read-only by design, masked account data only — runs daily checks on spend, cash position, recurring charges, and investments. Paired with market data for portfolio analysis and a Sheets connector with guarded writes: an agent can log a transaction but can never corrupt the ledger.
Health
A custom WHOOP connector tracks recovery, sleep, HRV, and strain against my own baselines, combined with a PubMed connector for agentic research support — including an 18-week marathon training block where AI plans each week from live recovery data.
Systems
Agentic macOS security diagnostics and remediation on my own machines — firewall hardening, permission audits, credential hygiene — plus agentic coding workflows that let me ship production software as a team of one.
How I Work
Three disciplines separate this from off-the-shelf AI adoption.
Codified knowledge
Every system ships with living knowledge bases that record how the platform works and every lesson learned — so the AI performs like a veteran employee from minute one and quality compounds instead of resetting.
Layered automation
Connectors give AI governed access to your data; agents run on schedules; briefings surface only what needs a human decision. People stop doing data entry and start doing judgment.
Safety by architecture
Read-only by default, human confirmation gates on anything consequential, kill switches, and no sensitive data in AI surfaces. Built to HIPAA-grade discipline, because I learned in the hardest industry for it.
The ROI logic is simple: fewer hours on documentation, follow-up, reporting, and reconciliation; faster lead response and higher conversion; revenue that stops leaking through unbilled claims and stockouts; and an owner who sees the whole business every morning without asking anyone for a report.
Contact
Working with me
I take on a small number of engagements with owner-led businesses: an AI operations audit to map where agents pay for themselves fastest, followed by build-and-operate work adapted to your stack. If your business runs on people remembering things, I can make it run on systems instead.
Book an intro call