Reno, NV · 14 years engineering, now building applied AI
About
I spent 14 years in developer relations and network automation at Cisco: writing NSO
service templates and YANG models, debugging live network state, and teaching other
engineers how to automate infrastructure they used to configure by hand. That's not a
resume line I'm stretching to sound relevant here. It's the actual foundation this
portfolio sits on: the discipline of building things that have to keep running once you
stop watching them, and the ability to explain how they work to whoever operates them next.
In 2026 I pointed that same discipline at applied AI. The projects under
Work are the result: a multi-agent research system, a RAG and MCP
backend behind a live product, a mobile app my own household depends on, a SaaS product
with real payments and a real security audit, and a workflow-automation business built and
registered as a Nevada LLC. All shipped, all reviewed line by line, none of them a demo.
How I build
I choose stacks by what the problem needs, not by habit. React Native kept a path open to
mobile without maintaining a separate native codebase. Python fit the data and agent work.
SvelteKit's server-first routing put every auth and content-access gate at 1 layer instead
of scattering client-side checks across the app. I'm not loyal to a language. I'm loyal to
shipping the right thing well.
Spec-driven, AI-assisted development is how I deliver production-grade work in whatever
stack a task calls for. I write every spec, review every diff, and make the calls that don't
show up in a diff: OpenTofu over Terraform for a licensing reason, killing a fully-built
payment integration once the data said it wasn't working, moving a token deduction before
generation instead of after. Claude Code writes most of the implementation and the tests.
That division of labor is disclosed on every showcase page, not hidden behind it.
How I operate
I don't just write code. I ship and operate real systems: infrastructure, CI/CD, payments,
security hardening, deployment, and observability. NovelFlame runs on infrastructure I
provisioned with OpenTofu, deploys through a CI pipeline I built with required status
checks and Docker layer caching, and survived a security audit that caught 4 real financial
bugs, including a payment webhook silently failing to credit real customers, before any of
them caused real harm. Family App runs a 3-provider AI failover chain and a 15-minute
health-check cron so a silent failure gets caught fast instead of running for days
unnoticed, which is exactly what happened to the WhatsApp bot it replaced.
I instrument the products, not just ship them. PostHog captures the real user journey in
both NovelFlame and MTG Commander AI: NovelFlame's story-creation funnel from first prompt
to a regeneration cap, MTG Commander's deck-analysis views, so I can see where people drop
off instead of guessing. NovelFlame's lifecycle and newsletter email runs through Loops.so,
triggered off those same product events.
None of these products carry a large user base, and I'm not going to pretend otherwise.
NovelFlame had 0 paying users at the point its billing model got reset. MTG Commander AI
runs on 1 Patreon patron. Sierra Story Co. has 1 completed memoir and no paying clients yet.
The engineering is the point here, not the traction: this is the discipline I bring when
I'm the only one checking whether something is actually working in production.
Skills matrix
A domain map of what I've actually built, not a language-fluency ranking. Every line links
to the showcase that proves it.
AI / Agentic
Multi-agent systems, retrieval, tool servers, and the guardrails that keep them honest.
- Multi-agent orchestration (parallel personas + deterministic judge) Finance Agent
- RAG pipeline design (chunking, embeddings, pgvector retrieval) MTG Commander AI
- MCP tool servers (FastMCP, usage-audited tool routing) MTG Commander AI
- Multi-provider LLM routing and failover (Anthropic, OpenAI, DeepSeek, Gemini, Groq, xAI) Finance Agent Family App NovelFlame
- LLM-as-judge evals and hallucination detection (batch validation against ground truth) MTG Commander AI Finance Agent
- Agent memory systems (dual-scope, per-user and per-household) Family App
- AI safety guardrails (defense-in-depth, static analysis + runtime allowlists) Finance Agent
- AI content-safety filtering (moderation API + custom rules, tiered response) NovelFlame
Backend / Data
Services, schemas, and the data-quality work that makes an AI system trustworthy.
- Python service design (typed agent loops, systemd-supervised daemons) Finance Agent
- Postgres + pgvector data modeling at scale (33K+ embedded records) MTG Commander AI
- Node.js / TypeScript backends with tool-calling agent loops Family App
- SQLite (WAL mode) for local-first, single-writer systems Finance Agent
- Row-level security and multi-tenant auth (Supabase Auth + RLS) Family App MTG Commander AI NovelFlame
- Batch data-quality pipelines (LLM-validated cleanup across tens of thousands of records) MTG Commander AI
- Schema-as-code (Drizzle ORM, version-controlled migrations) NovelFlame
Infra / DevOps
Infrastructure as code, CI/CD, and the observability that catches failures before a user does.
- Infrastructure as Code (OpenTofu, encrypted state, SOPS/age-encrypted secrets) NovelFlame
- CI/CD pipeline design (required status checks gating merges to main) NovelFlame
- Containerization (multi-stage Docker builds, layer-cached CI, GHCR) NovelFlame
- Cloud deployment (Railway, Vercel, Cloudflare R2/CDN/Turnstile) MTG Commander AI Family App NovelFlame
- Observability (structured logging, health-check crons, circuit breakers) Family App
- Self-hosted operations (systemd process supervision, LAN-only home-server deployments) Finance Agent Sierra Story Co
Frontend / Mobile
Native mobile, server-rendered web, and picking the platform the problem actually needs.
- React Native / Expo mobile app development Family App
- Native iOS integration (Swift App Intents, Siri, EventKit) Family App
- SvelteKit full-stack apps (server-first routing, SSE streaming) MTG Commander AI NovelFlame
- Self-contained, accessible UI systems (no external hosts, light/dark aware) This portfolio
Product / Operations
Payments, security, and the operator discipline of running something end to end.
- Payments integration (Stripe, Apple IAP, a full payment-processor migration and retirement) NovelFlame
- Security auditing and remediation (race conditions, timing-safe comparisons, webhook integrity) NovelFlame
- Data-driven product decisions (killing a shipped feature once the evidence said to) NovelFlame
- Small-business operations (LLC formation, service agreements, client delivery) Sierra Story Co
- Vendor-preflight QA automation (catching rejections before the vendor does) Sierra Story Co
DevRel / Communication
14 years of explaining complex systems to people who have to operate them, now applied to AI.
- Spec-driven development facilitation (writing every spec, reviewing every diff, directing AI-assisted implementation across 400-plus specs) Finance Agent MTG Commander AI NovelFlame Family App Sierra Story Co
- Workflow automation design (n8n pipelines, sandboxed-runtime workarounds, documented runbooks) Sierra Story Co
- Developer relations and technical enablement 14 years, Cisco DevRel
- Technical writing and documentation-as-code 14 years, Cisco DevRelThis portfolio
- Teaching complex systems to non-specialist audiences 14 years, Cisco DevRel