Building AI
Systems //
that Ship Revenue.
For enterprise teams and a small number of consulting engagements.
"I build AI systems that ship and operate. Not demos, not proofs of concept. Production infrastructure, deployed into enterprise commerce environments, measured by revenue."
Selected
Work.
28% Lift: AI Personalization System
FeaturedProduction deployment of real-time AI personalization across a large-catalog commerce environment. Built a system using vector embeddings to adapt content, product ranking, and messaging in under 200ms based on inferred visitor intent. Up to 28% CVR lift in tested cohorts.
hagl.co: AI-Adaptive Portfolio
FeaturedThis site is itself proof of work. Real-time visitor profiling, GPT-4o personalization, and pgvector semantic search running on Vercel Edge, delivering an adaptive experience to every visitor. It's the same kind of system I build for clients.
Post-Purchase AI: $139K/Mo Attributed Revenue
FeaturedLaunched a post-purchase communication system from zero and scaled it to ~$139K in attributed monthly revenue. Hit breakeven early ($54K paid / $56K made), then compounded. 67% open rates and 24% click rates portfolio-wide. Revenue figure reflects the attributed measurement window.
FY25 Experimentation Engine: 57 Tests
Designed and operated a full-year A/B testing program across 4 brands. 57 tests shipped, 34 positive revenue outcomes (60% win rate). Top single test: $606K projected incremental revenue. Consistent finding: value messaging and social proof outperform navigation and interface changes.
$2M Growth: Wellness Brand Scale
End-to-end digital infrastructure for a 0 to $2M build. Shopify Plus, high-conversion headless UI, and a retention program that grew email from 0% to 34% of total revenue in 18 months.
40% Revenue: Retention Architecture
Overhauled a stagnant email program with behavioral triggers, advanced segmentation, and AI-led content testing. Took email from a cost center to 40% of total brand revenue. 42% average open rates, $1.84 revenue per send.
Review Velocity: 800 to 2,700/Week
Feed-level integration fix that unlocked a compounding social proof engine. Review velocity went from ~800 to ~2,700 per week. 53K+ reviews surfaced on-site.
Personal Chief-of-Staff AI
Most AI assistants wait to be asked. This one monitors. A private, Telegram-native AI agent that runs a proactive heartbeat every 30 minutes: checks Gmail, flags calendar conflicts, surfaces action items, and delivers a concise digest. No UI required. Telegram is the command center, notification system, and interaction layer. Self-hosted on Hetzner, Python orchestration, Obsidian-style markdown memory vault.
AI Briefing Feed for Technical Founders
Solo AI builders and technical founders monitor too many fragmented sources to stay current. This replaces that habit with one daily touchpoint: a swipeable card feed where every story is framed as 'what this means for your work,' not just a summary of what happened. AI pipeline de-duplicates across sources and writes builder-specific impact bullets. MVP in progress.
Inbox as Podcast: Audio-First Email Player
Founders with 100–400 emails/day and no screen time. This connects to Gmail/Outlook, summarizes each message with GPT-4o-mini, converts to audio via OpenAI TTS, and queues it like a podcast with lock-screen controls. The goal: clear your inbox on your morning commute without touching your phone. PWA, $39/month flat, 14-day trial. MVP in progress.
AI-Powered Builder Progress Tracker
Builders lose momentum because logging progress is high-friction and private. This lets you log updates by voice or text. AI parses them, extracts metrics, and posts structured entries to a public timeline. The social graph follows projects, not people. The accountability mechanism is baked in. Slack integration, AI check-in nudges, public shareable pages. MVP in progress.
Not a manager
who builds.
A builder who leads.
I build AI systems: RAG pipelines, personalization engines, agentic workflows, semantic search. Built to run in production and move numbers. Not demos. Not proofs of concept.
Ten years in ecommerce is what makes the AI work different. I know what moves conversion, where the data lives, and how the customer experience has to feel at scale. That context is what separates a system that works in a sandbox from one that runs in production.
Currently Sr. Manager of Digital Experience at a large public safety enterprise. I own digital P&L across 4 brands, run the experimentation program, and lead a platform migration while keeping production revenue stable. Before that I ran Hagl Co., helping mid-market DTC brands build the revenue infrastructure they couldn’t hire for full-time.
A System that
Adapts to You.
(In Real Time.)
This portfolio is a live demo of edge personalization. It uses GPT-4o, pgvector, and Vercel Edge to adapt the experience for every visitor. It's the same kind of system I build for clients, running in production here.
Source Intent
Reads your referrer, location, device type, and time of day to get a read on who you likely are and why you're here.
Contextual Processing
GPT-4o takes that signal and decides which work to surface, how to frame the headline, and what context is actually relevant to you.
Adaptive Interface
Headlines, featured projects, and key details shift in real time at the edge. Under 200ms, no page reload.
Substantive Logic
Every project and skill is indexed in a pgvector database. The AI chat on this site uses that index to answer questions accurately.