AI systems & evals · Pacific time · remote or on-site

Agentic systems, shipped — and proven in public.

Multi-agent workflows built on Cloudflare, n8n and Claude — then run against a frozen test set with adversarial cases, behind a human-approval gate and a kill-switch. The failures stay on the page.

6
frozen eval cases
3
agents under test
2
adversarial traps
0
failures hidden
What I do

Three things, end to end

01

Build the system

Multi-agent workflows that do real work — plan, fan out, verify, synthesize. Wired on n8n, Cloudflare Workers, and Claude, and shipped end to end rather than demoed.

02

Prove it works

A frozen test set against the real system: deterministic checks plus a judge that must quote actual output or the score is thrown out. Adversarial cases included. The failures stay on the page.

03

Operate it behind a gate

Human approval on every outbound action, a global kill-switch, and fail-closed auth — so the thing can go in front of customers without anyone losing sleep.

Selected work

Case studies

Systems I've built and shipped — every one of them live and open to poke. Happy to walk through any of them.

Also shipped

More builds

Automation library

13 workflows, wired and shipped

The actual n8n flows behind the case studies — clinic reminders, an AI SMS qualifier with memory, error monitoring, and multi-source scanners. Each one is an interactive, importable canvas.

5 Clinic & appointments4 Missed-call & AI SMS2 Ops & reliability2 Alerts & multi-tenant
Browse the library →
How I build

My toolkit

Tool-agnostic and AI-assisted. I reach for the fastest reliable tool for the job — no-code where it fits, code where it earns its keep.

Evaluation & measurement
Frozen test setsDeterministic checksAnchored LLM judgesAdversarial / trap casesRegression tracking (D1)Uptime probes & ops records
Agentic systems
ClaudeGPTGrokMulti-agent orchestrationPrompt engineeringApproval gates & kill-switchesCloudflare Workers · D1 · KV
Automation & no-code
n8nMake (Integromat)ZapierAirtableSheets / Apps ScriptWebhooksREST APIs
Interfaces & delivery
v0FramerLovableReal-time dashboardsTwilio (SMS / WhatsApp)Deliverability (SPF·DKIM·DMARC)
Let's talk

Have an agent in production? Let's find out if it works.

Send a note — I'll reply within a day. Happy to walk through any system on this page live, including the parts that score badly.

or reach me directly

Pacific time · remote or on-site · EN / RU