I build AI-powered operational systems that raise both the quality ceiling and speed of delivery.
See What I've Built"Where architectural decisions matter: LLMs versus deterministic code, versioning for cost/quality tradeoffs, guard rails that catch edge cases."
Years Tech-Stacking Ops
Now Powered by AI
AI didn't change what I do. It changed how well I can do it. These are production systems I designed and built, not demos or experiments. Each one solves a real operational problem with measurable outcomes.
Iterated through 5 architectures (single agent, thin orchestrator with round managers, pipelined headless subprocess agents), gaining full observability into cost, quality, and concurrency tradeoffs. Pipeline state management with optimistic locking for concurrent sessions.
Stack: Python + Claude (Opus/Sonnet) + Replicate FLUX + Airtable state management
AI data normalization prototype mapping 25 fields across inconsistent vendor schemas. 15 resolved instantly via registry, 10 via Claude API with confidence scoring (7 fields at 95-100%, 3 correctly returned null). Demonstrated production-ready accuracy for enterprise ETL pipeline.
Stack: Astro (SSR) + Claude Sonnet API + Tailwind + Handsontable CE + Netlify
Shipped to market: multi-API orchestration (Replicate FLUX for generation, background removal for AR overlay, RevenueCat + Stripe for automated payment tracking), solving the integration challenge of making 4 independent APIs behave as one product experience.
Stack: Replicate FLUX + RevenueCat + Stripe + Airtable + Supabase Edge Functions
Complete automation loop from job discovery to post-application triage. Pre-application: pulls new postings, scores against 87-point rubric, routes structured evaluations. Post-application: Claude Chrome Extension triages inbox responses automatically.
Stack: n8n + Perplexity AI + Claude Chrome Extension + Airtable
Built standalone MCP server exposing proprietary tools to AI assistants. Among first third-party developers through OpenAI's App Directory process. Interactive widget with checklist tracking, budget calculator, and affiliate integration.
Stack: Node.js + Model Context Protocol + Render deployment
MCP server enabling two humans to share live project context across independent Claude instances. Async handoffs, shared task tracking, append-only decision logs, and key-value context store. Built for real joint ventures, not demos.
Stack: Python + FastMCP v2 + Supabase PostgreSQL
Multi-API orchestration producing marketing videos at $0.02/video with consistent quality at scale. Four APIs coordinated in sequence, each handling a discrete production step.
Stack: Claude + Replicate + ElevenLabs + ffmpeg
Custom Claude Code architecture with hooks, skills, MCP integrations, and cross-session memory that compounds output quality across every work session. iPhone capture syncs thoughts to active sessions in real time.
Stack: Custom skills + Apple Notes MCP + Airtable logging + Slack notifications + project memory system
58+ hours/week in AI-driven productivity gains with measurably higher output quality.
The same pattern across a decade: find the bottleneck, select the technology, build the system, train the team, measure the impact. AI is the latest (and most powerful) layer.
Match AI to task like you would talent: deploy where it exceeds humans (pattern recognition, consistency, scale) to increase bandwidth for work humans do best. Production systems require rigor: strategic LLM usage vs. deterministic code, cost observability, ROI validation.
Production AI requires skepticism. I version architecture to find cost/quality/concurrency sweet spots, use LLMs only where they beat deterministic code (dollars to pennies), and build guard rails: confidence scoring on outputs, null handling for ambiguous cases, cost tracking per operation. Python orchestration with thin coordinators routing specialist agents—not monolithic black boxes.
Training operations for 350+ BMW dealerships were fragmented across spreadsheets with no visibility, inconsistent processes, and mounting overhead.
Evaluated, selected, and drove org-wide adoption of an integrated 5-platform tech stack spanning payments, automation, project management, and collaboration. Built automation workflows that cut overhead 50% and were replicated org-wide as the standard. Built analytics infrastructure enabling 20%+ margin improvements through data-informed stakeholder engagement.
New product needed rapid feedback systems and scalable QA processes to iterate on real user behavior.
Designed and implemented feedback capture systems producing 50% increase in customer data within 1 month. Led implementation of scalable QA processes achieving 30% reduction in testing cycle time within 6 months.
Legacy breeding facility trapped the owner in day-to-day scheduling, tracking, and logistics. All operations ran on manual record-keeping.
Re-engineered from manual record-keeping to integrated digital systems. Modernized scheduling, animal tracking, logistics coordination, and reporting.
Six field reps managing 10-30 simultaneous productions with no centralized tracking. Contract enforcement inconsistent, compliance documentation scattered.
Designed a multi-stakeholder data system that served compliance, field ops, auditors, and federal agencies simultaneously. One system, five audiences. Introduced Slack and digital coordination in an organization running on manual processes. Leveraged analytics to deliver first contract improvements in 15+ years.
High-profile event operations lacked standardized workflows. Scheduling accuracy and team communication were inconsistent.
Designed and implemented workflow and scheduling systems for 100+ high-profile events. The first iteration of a career-long pattern of building technology-driven operational infrastructure.
Production AI Systems · 350+ Sites · Multi-Million Operations · Media, Auto, Fintech, Sports
I've built operational systems across media, automotive, fintech, and sports since 2012. The pattern doesn't change: find the bottleneck, architect the system, scale it. Whether it's multi-stakeholder platforms serving compliance and field ops simultaneously, or payment systems orchestrating across multiple rails, the discipline is the same: architect for the problem, scale for the organization.
Architectural decisions determine whether AI drops costs from dollars to pennies or becomes another expensive black box. I version architectures to find cost/quality sweet spots—single-agent systems, thin orchestrators, pipelined workflows with full observability. Evaluation systems with confidence scoring. Multi-API orchestration with null handling and per-operation cost tracking.
Production systems require skepticism, not faith. AI genuinely exceeds human capability across many domains. Production rigor means knowing when AI is the right tool and when deterministic code wins—then deploying with guard rails, not blind trust.
AI didn't change what I do. It changed the ceiling on quality and speed when you build with rigor instead of hype.
"Rigor instead of hype."
Role: COO | VP of Operations | VP of AI & Automation | Head of AI Operations | Chief AI Officer | Director of Business Intelligence
Company: Profitable, AI-forward company where building systems that actually work matters more than talking about transformation
Location: Denver, CO · Open to remote
Industries: AI-forward tech/SaaS, AI-native companies, or established companies needing technology-forward operational transformation