A working
architect,
on file.
AI-native systems, written down.
I design and ship AI-native enterprise systems for teams that need the hidden 90 percent to work in production. Multi-agent orchestration, async queues, cost-aware deployments, the full plumbing.
What follows is a working file: case studies of systems I have shipped, with the constraints, decisions, and code where the claim lives.
Shellwire
An autonomous newsroom run by twelve LLM providers voting against each other.
pSEO Engine
A programmatic-SEO engine that ships static editions, ranks them, and ports the pattern to any vertical -- proved end-to-end on myphotoai.alkenacode.dev and agents.alkenacode.dev.
A2UI free
A free-tier deployment of Google's agent UI protocol, running on 92 live LLMs nobody pays for.
Aether
A local-first knowledge graph PWA — Obsidian alternative with daily Pulse reports refined by the same council pattern as Shellwire.
Optimise for the hidden 90.
API wrappers are not architecture. The interesting work lives in queues, fallbacks, rate limiters, idempotency, observability. I build for that layer first.
Cost belongs in the design, not the bill.
Free-tier matrices, Redis-backed quotas, fallback chains, request coalescing. A system that ignores inference cost gets shut down.
Every decision should be replayable.
Council votes logged with latency and tokens. Pipeline runs persisted to Redis. Outreach traced end to end. If you cannot replay it, you cannot improve it.
Distribution is part of the product.
Programmatic SEO, sitemap integrity, IndexNow, structured data. Not afterthoughts. The system that ships content also ships proof of the system.