Portfolio

Systems I built and run, client work described without naming clients, and the open-source tooling behind it.

Systems Built and Running

Support Forge

Support Forge

AI consultancy serving 20+ client organizations across manufacturing, healthcare, legal, real estate, and professional services. AI enablement engagements, agent systems, governed automation, and hands-on team training.

AI ConsultingEnablementAI AgentsTraining

Manufacturing CPQ Engine (coached build) + Whitepaper

Coached a building-products manufacturer's in-house developer to ship an AI-assisted configure-price-quote engine, now in a production parallel run against the legacy system, with guardrails, a Petri-based test plan, and behavioral regression checks. Co-authored the client-approved whitepaper on the six failure modes surfaced along the way.

ManufacturingCPQCoachingWhitepaper

Healthcare Operations Agent

Claude-based operations agent for a healthcare staffing organization, working under a signed BAA. Validated on a 104-document remittance analysis batch with draft-only email automation — a human approves everything that leaves the building.

HealthcareAI AgentsComplianceHuman-in-the-Loop

Ops & Finance Command Center

The system Support Forge runs on: eight agent roles with approval queues, trust scoring, and cost caps; live ingest from Stripe, QuickBooks, Gmail, Calendar, and Fathom; pipeline forecasting; 500+ automated tests; push-to-main deploys on Google Cloud Run.

Claude CodeCloud RunAgentsFinance Ops

Partner (BD Rep) Portal

PWA for referral reps with server-side scoped data projections, PIN auth with lockout, a commission ledger, and per-rep activity timelines, plus a headless proposal follow-up worker that verifies sent mail before it drafts and never sends.

PWAAuthCommissionsHeadless Agents

Alignment Audit → Runtime Safeguards (open source)

An 86-scenario Petri alignment audit of a production model, the safety supplement it produced, and the 3-layer runtime canary that enforces it (pre-action tripwire, self-audit, false-positive detector), plus a model-upgrade regression harness. Both repos public.

AlignmentPetriHooksOpen Source

Cert Forge

Exam simulator and study tracker for the four Anthropic partner certifications: 266-item question banks, spaced retrieval, magic-link auth, a scoreboard, and RLS-locked answer keys. Cloud Run + Supabase.

Next.jsSupabaseRLSEdTech

Interactive Prospect Demos

A noindex demo subdomain hosting interactive, no-PHI command-center demos built from each prospect's own discovery-call vocabulary: an O365 executive assistant, a home-health COO weekly review, a real-estate broker center, and more.

Firebase HostingDemosSales Engineering

Regional Planning Commission Agent System

Governed Claude Code agent system for a regional planning commission: written operating principles enforced live (the auth gate tripped on the first email attempt), per-role agents for grants, finance, and data, Microsoft 365 via desktop connector, and skill packs with hooks.

Public SectorGovernanceMicrosoft 365Claude Code

Field Services Integration Stack

Replaced a brittle Zapier setup for a field-services business with direct Jobber, QuickBooks Online, and dual-Gmail integrations orchestrated through Claude Code — fewer moving parts, no per-task fees, fully owned by the client.

IntegrationsQuickBooksAutomationSMB

Project Pipeline Dashboard

Unified kanban project dashboard for a manufacturing technology lead, mined from ticketing, SharePoint, and 90 days of email — surfacing projects that existed only in inboxes. Drag-and-drop hierarchy, progress gauges, and draft-only email actions.

DashboardsEmail MiningKanbanAI Agents

Consulting Services

AI Enablement Engagements

Fixed-scope, outcome-priced engagements: readiness assessment, use-case prioritization, the build, governance, and hands-on training until the owning team runs it without me.

EnablementTrainingFixed ScopeOwner-Operators to Mid-Size

Claude Code Agent Systems

Subagents, hooks, custom MCP servers, agent skills, and headless scheduled agents, designed so every action that leaves the building is a draft a human approves.

AI AgentsMCPClaude CodeHuman-in-the-Loop

Governance & Evaluation

Alignment testing with Anthropic's Petri framework, runtime canary hooks, and a model-upgrade regression harness so behavior is re-baselined whenever the model changes.

PetriCanary HooksRegression HarnessResponsible AI

Regulated-Industry Builds

HIPAA-bound work under a signed BAA, draft-only email, and data custody that stays with the client. Built for home-health, legal, and public-sector environments.

HIPAABAAData CustodyPublic Sector

Integrations That Replace Glue Tools

Direct API integrations into Stripe, QuickBooks, Jobber, Google Workspace, and Microsoft 365, orchestrated through Claude Code, replacing per-task Zapier plumbing with something the client owns.

QuickBooksStripeGoogle WorkspaceMicrosoft 365

Training & Speaking

Client training engagements, train-the-trainer programs, and public sessions: Haverhill AI Summit (June 2026) and "AI for the Rest of Us" at Boston AI Week (September 2026).

TrainingTrain-the-TrainerWorkshopsTalks

Ready to Transform Your Business with AI?

Let's talk about which workflow goes first, what governs it, and who on your team runs it after handoff.

Get in Touch