← Back to Home

Writing

Architecture write-ups on building production AI systems on Cloudflare.

Failure Under Load: Fixing a Race Condition and Adding AI Fallbacks

Jul 17, 2026

How I hardened SE Intel against two production failure modes: a KV rate limiter that lost counts under concurrency, and Workers AI failures that needed a 70B → 8B → canned-response fallback chain.

cloudflareworkersdurable-objectsworkers-aireliabilityfallbacksrate-limitingse-intel

How I Built a CI Gate for My AI Agent (And It Caught a Real Bug on Day One)

Jun 25, 2026

How I added a deterministic faithfulness check and LLM-as-judge eval pipeline to SE Intel — blocking deploys on quality drops and catching a real flaky bug on the first run.

cloudflareworkers-aievalscifaithfulnessllm-as-judgeai-engineeringse-intel

Multi-Tenant Isolation in an Edge AI System — Three Layers, Three Probes, Zero Trust

Jun 18, 2026

How I isolated RAG, memory, and audit data by organization in a multi-agent AI platform on Cloudflare Workers. Every isolation boundary is proven by a deterministic probe — no LLM in the test path.

cloudflareworkersaimulti-tenancyisolationdurable-objectsvectorizesecurity

How I Built a Role-Gated Multi-Agent Sales Intelligence Platform on Cloudflare Workers

Jun 2, 2026

Every architecture decision behind SE Intel — Durable Objects for agent state, three-tier RAG with namespace filtering, streaming SSE, LTM extraction, and an LLM-as-judge eval harness.

cloudflareworkersaidurable-objectsvectorizerag