Architecture / Operating systems for intelligence

Make the
whole system
think.

AI runtimes, model operations, identity, and APIs designed as one production architecture.

Live architecture

One mind.
Three operating planes.

Select a system and watch the active path redraw around the engineering intent.

RUN / 01 SYSTEM ONLINE

Orchestrate judgment.

Planner, specialists, scoped tools, eval gate.
INPUTPLANNERRESEARCHCRITICEVALACTION

Selected systems

Architecture is the work.

Proprietary details stay private. The decisions and their consequences do not.

01

Applied AI / Multi-agent systems

Automation that exercises judgment.

Problem
Recurring analysis and content workflows needed multi-step decisions, not another brittle scheduled script.
Decision
Planner and specialist agent topologies with scoped MCP tools, n8n orchestration, and an evaluation gate before external action.
Impact
Auditable systems that plan, act, critique, and revise their own work on schedule.
CRON / EVENTPLANNERMCP TOOLSEVALSAUDIT LOG
02

MLOps / Model operations

Models that notice when the world changes.

Problem
Production models degrade quietly as live data moves away from the distributions they learned.
Decision
PSI and KS drift monitoring, automated retraining in CI, versioned models, and explicit promotion gates.
Impact
Model updates become evidence-led engineering events instead of emergency fixes.
LIVE TRAFFICDRIFT · PSI/KSRETRAIN · CIREGISTRY vN+1
03

Enterprise / Identity

One identity layer. Millions of users.

Problem
Fragmented login flows made every channel inconsistent and every migration risky.
Decision
A phased move to a central Okta platform using OAuth 2.0 and OIDC without interrupting daily traffic.
Impact
A shared trust layer across web, mobile, and partner channels.
WEBMOBILEPARTNERIDP · OIDCTOKEN → APIS
04

Enterprise / API platforms

A typed contract behind every profile.

Problem
Tightly coupled services turned every profile change into a shared deployment bottleneck.
Decision
NestJS and TypeScript microservices behind GraphQL, with Couchbase and deliberate caching tiers.
Impact
Independent teams, predictable contracts, and low-latency reads at enterprise scale.
CLIENTSGRAPHQL GWPROFILE SVCPREFS SVCCACHE · KV

Capabilities / In order of focus

01

Applied AI
& Agents

Multi-agent systems with MCP. Claude, GPT, Gemini, and OpenRouter integration. RAG, tool use, prompt engineering, evals, and n8n orchestration.

02

MLOps
& DevOps

Training, model versioning, drift monitoring, automated retraining, Docker, CI/CD, GitHub Actions, observability, AWS, and Vercel.

03

Engineering
& Architecture

Okta, OAuth/OIDC, NestJS, TypeScript, GraphQL, REST, Couchbase, microservices, and high-traffic identity and profile platforms.

Next system

Enter the AI lab