AI in production. Inside Drupal sites, and on its own.
PillarShield is a live Drupal module, a live WordPress plugin, and a REST API at api.pillarshield.co. It does real-time AI content governance at the publish boundary. I designed it, built it, and run it. Nine other non-Drupal AI tools for local business at pillarpoint.consulting.
Live Drupal module. Live WordPress plugin. A REST API any CMS can call. Built and operated by one engineer.
$ curl -X POST https://api.pillarshield.co/pillarshield-governance/governance \ -H "Content-Type: application/json" \ -d @publish-event.json HTTP/2 200 OK { "event_id": "1c9376fb-b0bf-44e5-bfe0-1702fe55ba37", "decision": "block", "status": "VIOLATION_DETECTED", "plan": "protect", "violations": [ { "type": "PII_INDICATOR_OTHER", "severity": "error", "message": "Student PII (names, GPA, financial aid)." }, { "type": "PII_INDICATOR_OTHER", "severity": "error", "message": "HIPAA medical info." }, /* 3 more violations elided */ ], "meta": { "content": { "url": "pillarshield.co/blog/.../150-authors-one-publish-button-zero-safety-net" }, "benchmark_ms": { "total": 1483.74 } }, "chain_hash": "be909c3e03f35daff3cae8db182caeeec2161f1e41c0b3944557be95b51e9a9e", "previous_chain_hash": "bc8b35d6fdc1c96c4296d18199c8352a15923ad4192e77cde6da5c70f9000cbe" }
The workstation I sit at every day is a local-first, multi-modal, multi-model running machine, with the heavy reasoning farmed out to the cloud only when the problem is worth the round trip.
48 GB of VRAM. Enough to run a 35B-parameter model locally at usable speed, without sending a token off the machine.
LM Studio is the daily driver on Rune. Ollama serves the worker laptops. vLLM stays in rotation for higher-throughput batch work.
My default local model. Fast enough on the rig, capable enough for almost every refactor, code review, and architecture sketch I throw at it.
For the work that needs deeper thinking: gnarly debugging, large refactors, architectural exploration across PHP and Python.
Ties the local models, Claude, and the surrounding tools together. Memory injection, RAG against Qdrant, streaming proxies that make multi-model workflows feel native.
The part the models can’t do. Writing production software since 1989. I know when the AI is wrong, and I know when to override it.
Lead backend engineer on Smartsheet’s Drupal-based marketing, help, and community platform. Architected the original Coveo-based search, then led the strategic migration end to end.
When I started at Smartsheet, I was tasked with designing Smartsheet’s enterprise federated search on Coveo across marketing, help, and community properties. Then I led the migration to OpenSearch on AWS: cluster design, index and document mapping, query DSL, BM25 and function scoring, relevance tuning, and production cutover. The licensing cost went to zero. The customer-facing search quality did not move.
I’m crashtest_ on drupal.org. Credited core contributor. Author of two modules, Fences and Key, that are now standard infrastructure on a large fraction of production Drupal sites. This is where I’ve lived professionally since February 2008.
The condensed view. The full timeline lives in the résumé.
Founded the practice. Built PillarShield, the live Drupal module, WordPress plugin, and REST API for AI content governance. Self-hosted the AI infrastructure that powers internal development. Daily user of local LLMs and Claude Code as first-class engineering tools across PHP and Python.
Lead backend engineer for Smartsheet’s Drupal-based marketing, help, and community properties. Architected Coveo search, then led the migration to OpenSearch on AWS that saved Smartsheet seven figures on search. Integrated Salesforce, Cvent, Greenhouse, YouTube, and Higher Logic Vanilla into the Drupal platform. Modernized legacy PHP into Symfony services.
Architected a multi-tenant Drupal platform and central installation profile consolidating 120+ independent Forest Service sites under a maintainable core. Led the Drupal 7 to 8 migration for the HQ site. Advised the CIO’s office on workflow, testing strategy, and long-term platform direction.
Built and maintained headless Drupal services powering NBCNews.com and Today.com editorial tools. Custom REST endpoints and caching strategies that supported millions of daily page requests.
Full-stack Drupal for the Chicago Field Museum, Riot Games (League of Legends), the Bill & Melinda Gates Foundation, Foreign Affairs, The King Center, Acana Pet Foods, Intuit Quickbase, Shin-Etsu Silicones, and the official site for Lady A.
Every one of these was a Drupal engagement. Every one had real users and real consequences when something broke.
An honest accounting, not an aspirational list. These are the tools I’ve shipped production work in within the last two years.
If you’re a Drupal shop or product company adopting AI, or an agency that needs an engineer who can ship AI features into a Drupal platform in production, this is the lane I want to be in. Drop a note and I’ll get back inside a business day.