SHIPPED SYSTEMS · FILE 006
Every system on this page is real: built by me, shipped to real users, and operated in production. No labs, no demos, no borrowed case studies. The AI work sits on top of over a decade of building the platforms my own businesses ran on.
Live
App Store Product
8,000+
Embedded Research Chunks
33
Behavioral Analysis Capabilities
3,000+
Entrepreneurs Assessed
AI SYSTEMS · IN PRODUCTION NOW
SYSTEM 01
LIVE · APP STORE
A behavioral intelligence platform for couples I built end to end and operate in production. A three-stage RAG pipeline over 8,000+ embedded research chunks, 33 behavioral analysis capabilities, voice AI, and production eval gates that decide what ships. Product, backend, payments, the iOS release pipeline, and deploys: all mine.
RAG architecture · Eval gates · Voice AI · Payments · iOS shipping
SYSTEM 02
LIVE · CLIENT PRODUCTION
Nette AI answers the members of a 200+ person mentorship so its founder doesn't have to. She expected to wake up to a full inbox every day. She wakes up to about three messages, and they aren't support questions. That gave her back the hours to launch her mentorship program and scale her reach, while the paid membership it serves renews month after month and its usage data feeds her retention and product decisions. Under the hood: her own training library embedded for retrieval (4,581 chunks), a lean rebuild that cut answers from timeouts to about 23 seconds, and me operating it like an SRE.
Business outcomes · Agent design · RAG on client knowledge · Production ops
“We have over 200 people in our mentorship. I was expecting so many people to be inboxing me day to day. I may have three messages, and they're not about group homes. Nette AI has freed up my time.”
SYSTEM 03
IN PILOT · FIRST COACH LIVE
The platform version of what Nette AI proved: a coach's course library becomes a proactive AI coach that guides each student instead of waiting for the right question. Real-time voice coaching in the coach's own cloned voice, per-coach knowledge retrieval, and fail-closed multi-tenant security. Built so one expert can be present for an entire community at once, without hiring a support team.
Voice AI (real-time) · Multi-tenant architecture · Per-tenant RAG · Product design
SYSTEM 04
IN PRODUCTION
An AI coaching agent that lives at the edge: Cloudflare Workers with Durable Objects holding per-user state, versioned database migrations, and an automated test suite. Built for instant cold starts and zero servers to babysit.
Edge compute · Durable Objects · Stateful agents · TypeScript
SYSTEM 05
DEPLOYED
A Python backend that makes LLM products affordable to run: multi-layer caching with a semantic layer that recognizes similar questions before they ever reach the model, multi-model fallback, deterministic state detection that runs before generation, and streaming responses. Cost engineering is a feature, not an afterthought.
Python + FastAPI · Semantic caching · Multi-model fallback · Streaming
SYSTEM 06
OPERATING DAILY
The control plane for the agent fleet that runs my businesses: launch, watch live, and review work across multiple coding and operations agents from one cockpit, with streamed output and a persistent run history. When the AI page says agent fleets run my businesses while I sleep, this is the system it means.
Fleet orchestration · Live streaming UIs · Run persistence
SYSTEM 07
BUILT · PUBLIC CODE
Eight specialized agents that walk a business owner through offers, money models, and revenue decisions, each grounded in published business frameworks and orchestrated inside a single application. Multi-agent systems only work when each agent knows exactly what it owns.
Multi-agent orchestration · Prompt architecture · Next.js
SYSTEM 08
DELIVERED · CLIENT SYSTEM
A complete AI spokesperson system for a coach: cloned voice, an approved stylized avatar built from real photos, and lip-synced video generation. Five-minute videos at a small fraction of filming cost, with a proven per-video cost model.
Voice cloning · Avatar pipeline · Cost modeling
THE DECADE BEFORE · WHY THE AI WORK SHIPS
Agentic AI did not teach me to build products. My own businesses did. Before the AI wave, I designed, built, and operated the platforms they ran on.
SYSTEM 09
OPERATED FOR YEARS
The platform my coaching business ran on, built from scratch: a client CRM with per-client detail and coach performance views, a course and video learning portal with an admin CMS, an assessment command center, a communication hub, calendar, membership, community, and analytics. One founder, one codebase, an entire back office.
CRM design · Learning portals · Admin CMS · Analytics
SYSTEM 10
OPERATED · PAID MEMBERS
A compliance training portal for a regulated residential-care niche: structured programs with phased lessons, a searchable compliance binder, an AI chat assistant over an admin-managed knowledge base, deal calculators, and subscription-gated access for a paid member community.
LMS architecture · Knowledge-base chat · Subscription gating
SYSTEM 11
BUILT · PUBLIC CODE
A form-builder platform built to run identity assessments at scale: an admin form builder with sliders, rankings, and uploads, public assessment links that need no signup, per-form analytics, and AI-generated personalized reports delivered through workflow automation. The lineage that leads straight to today's 12-question diagnostic.
Form-builder systems · Workflow automation · AI report generation
SYSTEM 12
YOU ARE READING IT
A guarded design system where every page's copy lives in tested content modules, truth-guard tests block invented numbers and confidential names from ever shipping, and every deploy is verified live before it counts. On this site, honesty is enforced by the test suite.
Design systems · Truth-guard testing · Next.js
WHAT THIS ADDS UP TO
Production agent design, retrieval pipelines, vector search, evals, and the guardrails that keep them honest.
Caching layers, model fallbacks, streaming, and the engineering that keeps an AI product's unit economics sane.
Web, mobile, and backend: from CRM and learning portals to an App Store release pipeline, built and operated solo.
The other half of the career: coaching, workshops, and behavioral frameworks that turn skeptical teams into daily users.
This is what I do: build agentic AI that survives contact with production, and get the humans on the other side to actually adopt it. Start the conversation.
