AI from prototype to production, with the guardrails built in.

• Payment-to-provision checkoutDelivery, AI, Rapid Prototyping23 steps to 1 click; v1 live in 5 days; zero downtime through my last day

  A self-service path for low-priced product lines. Discovery with the VP of RevOps, the CRO, and the product SME separated the steps the business needed from the ones the manual process had accreted; the target flow ran across a checkout front end, an application database, Stripe for payments, n8n for the handoffs, and Salesforce as the system of record for sales and finance. Payment events triggered the workflows that wrote the records and started provisioning. Two product lines and six payment tiers in production; the company’s first revenue-producing production application built with AI tooling; built with an AI agent harness writing the code, every line reviewed, supported in production after go-live.

• Compliance-document extraction pipelineAI, Governance$0.003 per record; 25 entities, 1,300 records; 26 percent cheaper at full fidelity

  A knowledge-graph-based LLM pipeline that converts 150-plus-page compliance filings into structured, evidence-scored records, replacing days of analyst review per company. Model-routed across premium and standard tiers through an LLM gateway; every run reproducible and version-controlled; evidence scoring on every extracted field; audit events logged per run.

• Governed application template and LLM gatewayGovernance, AI, Enablement, Rapid PrototypingGuardrails that warn, log, and inform; an org-wide template every department built on

  The reference architecture for AI-enabled internal applications: every AI call routed through a gateway with virtual keys and no raw provider keys; backend logic in edge functions only; multi-step orchestration and persistence in n8n; database-backed rate limits; PII redaction on by default; AI output written to a propose-then-approve table and never committed directly; audit logging on every function; a domain allowlist for outbound webhooks; AI-generated access policies reviewed by a person before deployment. Governance philosophy written into the template: enable people, do not block them; presentation state only in the frontend, business truth in the systems of record.

• Internal enablement platform and CoE libraryEnablement, Governance137 users onboarded; 97 percent adoption; about 1,400 hours a month recovered

  Built from scratch and onboarded every user: git-backed configuration, custom role-based access, API and MCP integrations, loop-based health checks, local and cloud database tiers, and a branded design system. The library carries 13 edge functions, 105 migrations, and 120 row-level security policies with a blocking type check and a passing test suite in CI. Authored the AI-literacy curriculum, the governance guardrails, and the reference architectures non-technical teams used to build and run their own workflows.

• Contract-validation workflows in productionDelivery, GovernanceLive about six months; Salesforce integration; PII controls; error handling

  Production n8n workflows that validate contract data against the CRM, with PII compliance built into the flow, a dedicated error-handling workflow, and a production checklist before each release; versioned through eight production iterations.

• Brand-governed content studioAI, EnablementLive with internal users; multi-format generation; brand voice enforced

  A content studio that generates LinkedIn, email, blog, and presentation drafts with the company’s brand voice enforced on every output; authentication and user management on the application database; the model called through edge functions behind the gateway.

• Security and governance reviewGovernance, Rapid Prototyping44 findings, 0 false positives; 1 critical PII leak remediated; 53 privileged functions hardened

  A comprehensive review of the platform’s edge functions and database: every privileged function mapped and hardened, a cross-repo single sign-on cutover with dual acceptance and zero downtime, and an org-wide AI governance policy for rapid prototyping adopted across departments.

• One auditable revenue data layerDelivery4 systems, 1 view

  Internal applications that consolidate four revenue systems (Salesforce, 6sense, HubSpot, Salesloft) into one traceable, auditable data layer for SDRs, account executives, and sales leadership.

• Agentic operating layer and skill ecosystemAI, Enablement30 components; validates its own configuration; architected for audit

  A skills-based operating layer that encodes each department’s domain knowledge as reusable, secure, scalable patterns on deterministic frameworks, with a PR-gated distribution repository, evaluation suites, and release runbooks so the next team starts from a pattern rather than a prompt.

The code is written with an AI agent harness, and every line is reviewed before release.

• frontier-operating-manual

  An operating discipline for AI coding agents, shipped as a Claude Code skill: standing rules with failure conditions, risk tiers, calibration labels, a written pre-send self-test.

  What it shows: governance as a runnable harness.

• ai-knowledge-repository

  A validated knowledge graph of hard platform constraints and decision-enabling capabilities for Claude Code, Lovable, Git, reMarkable, n8n, Python, and Atlassian, with a contract validator.

  What it shows: knowledge management built for decision time, with receipts.

• claude-code-swarm-kit

  A model-tiered multi-agent research kit: pinned roster, five-stage workflow, spawn doctrine, PreToolUse gates, and linters.

  What it shows: cost governance as code, root-cause work, honest verification records.

• reasoning-bundle

  A Claude Code plugin of structured-reasoning skills, opt-in calibration and convergence hooks, and a graph-of-thought tournament, with zero data egress.

  What it shows: packaging discipline as a distributable, adversarially reviewed artifact.

• remarkable-agent-push

  A security-audited, hardened build of a CLI that lets an AI agent push markdown to a reMarkable tablet as a native notebook.

  What it shows: source-level security audit, fail-closed fixes, a rebuilt release pipeline.