I build AI-integrated
automation systems
I design, automate, and ship from concept to production. Every system here is real and running, and I made the calls behind it: what code can do, what needs AI, and what stays human.
See the systemsSystems I've Built
Real systems with real users, not tutorial projects. I decide what to build, how it works, and where a human stays in the loop, and I build the automation layer myself. The application code is written by AI under my direction.
Agentic AI Workflow for Salesforce
An agentic AI system I designed from scratch that turns Claude Code into a consistent partner for Salesforce support: every ticket runs the same five-phase lifecycle, from discovery to a validated root cause. It ships a governance layer most AI tools skip: automated PII redaction, risk-tiered command safety, and a confidence-check that grades the evidence behind each finding and caps how certain the AI can sound before it reaches a stakeholder. A self-improving retrospective loop sharpens the skills after every closed ticket. Sixteen custom skills cut investigation time from hours to minutes across 98+ tickets, and after it proved out, leadership adopted it department-wide and tapped me to lead weekly training for the admin and developer teams.
NeuralTrace
AI memory that carries across tools. Save context once in Claude, ChatGPT, Gemini or Codex and recall it from any of the others. Chrome extension with semantic search, auto-enrichment, and a self-hosted MCP server.
Custom Booking & Payments System
End-to-end booking platform I designed and shipped for a sports and events photography client. Seven n8n workflows orchestrate slot reservations, payment capture through Square, and automated confirmations by email and SMS - including a 15-minute stale-slot reset that returns abandoned checkouts to inventory instead of letting dead carts hold sellable capacity. It ran the client’s live mini-session events end to end, on real customer payments.
bs-check: Confidence Audit for AI Agents
An open-source Claude Code skill that makes an AI prove its claims before you act on them. Two lanes: a reasoning audit that asks whether you are solving the right problem at all, and a verification pipeline that breaks a claim into sub-claims, grades the evidence behind each one, and returns a hard PASS or BLOCKED. Judges run in isolated context so the reasoning that produced a claim never gets to defend it.
Website Tracking Analyzer
Production SaaS tool that audits marketing tracking implementations across 25+ platforms. Detects Pardot, GA4, GTM misconfigurations and generates actionable diagnostics with root cause analysis, confidence scoring, and GDPR/CCPA compliance testing.
Time Tracking & Invoicing Platform
Full stack application, deployed and running: React and TypeScript client, Express and Prisma API across 13 data models, Docker and Traefik with automatic SSL. Every card carries a timer, every timer becomes a billable entry, and entries roll up into a client invoice with per client rates and tax. In the production instance, invoices post to an n8n pipeline that creates them in the accounting system; the public demo exports the same invoice as PDF or CSV. Audited against 15 competing tools, then shipped the fixes: concurrent session handling, security headers, rate limiting, and link validation.
What I Work With
Grouped by depth, not by logo count. Hands-on means I can be questioned on it without notes. AI-assisted build means I specified it, directed the implementation, and run it in production. I would rather tell you which is which than have you find out.
Salesforce & Marketing Ops
- Pardot / Account Engagement (10 yrs)
- Salesforce Administration (3 yrs)
- Declarative Automation (Flow)
- SOQL & Metadata
- Lifecycle & Campaign Ops
Automation & Integration
- n8n (50+ workflows)
- Webhooks & REST APIs
- Square, Twilio, Brevo
- Shopify, QuickBooks
- WordPress / Beaver Builder
AI Tooling
- Claude Code / API
- MCP Servers
- Agent Skill Design
- Prompt & Context Design
- Multi-Agent Verification
Infrastructure & Code
- Docker & Traefik
- Linux / VPS Operations
- Node.js
- Python
- Playwright
- Git / GitHub
Plan It. Research It. Then Build.
Most of the work happens before anything gets built. By the time the implementation starts, the hard decisions are already made.
Research
Challenge the brief, then pressure-test the approach
Plan
Map the flow end to end before anything gets touched
Build
Specify it precisely, then direct AI through it
Verify
Check the output where being wrong would cost something
Get in Touch
Want to talk shop about AI, automation, or infrastructure? I'm always up for a good conversation.