Blog
Field notes from operating private AI infrastructure, GPU inference, agent tooling, and healthcare integration systems. For longer implementation writeups, see Writing and Case Studies.
Showing all 30 posts
All Posts
The Factory Files a Ticket Against Itself: Two Repos, One Session, Three Gaps
I put fi-fhir and edilint through Mills for the first time. It merged work I did not write, then exposed a test gate passing on the wrong tree and fixed it against itself.
The Factory Upgrades Its Own Foundation: An MCP Spec Marathon
MCP shipped its first deliberately breaking spec revision. I wrote the gap analysis, filed it as factory backlog, and Mills started rebuilding the SDK it runs on.
The Agent Gets Read-Only Credentials: A Trust Ladder for AI in Regulated Systems
How I run LLM agents against regulated healthcare payment data: a three-tier trust ladder, kill-tests before promotion, and guardrails in infrastructure, not prompts.
A Green Run That Checks Nothing: Seven Silence Modes, Seventy Repos, One Day
A fleet-wide maintenance day exposed seven ways CI can report success without useful work—and the merge-queue discipline and recurring checks built to catch them.
The File That Failed With Zero Errors: A Field Guide to Invisible EDI Defects
Four defect classes that healthcare interchange parsers accept silently, with synthetic samples, real linter output, and the pre-send gate that catches them.
Twin Life on a Doubled Pool: One Day of Canary-Driven Inference Engineering
A twin-lane canary promoted prefix caching, FP8 KV cache, and a larger context pool while catching five regressions before they reached the primary workload.
Five Voices, 200K Tokens, One Consumer GPU: Giving Simulated Minds a Whole Life to Remember
How five simulated minds share one 24 GB Radeon, keep 200K-token life histories, reuse them through prefix caching, and compact old memories without erasing continuity.
Gaming Mode: Turning an Inference GPU Node into a Moonlight Host — Declaratively
I replaced FlexInfer’s unused gaming-mode stub with a declarative GamingSession resource that drains inference and turns a GPU node into a Moonlight host.
State of the Platform, July 2026
A July 2026 platform snapshot: multimodal inference, Mills delivery, flight recording, and the operational gaps that remained after the demos worked.
Finding the Real Context Ceiling: Needle-Benchmarking Forced RoPE Extrapolation
A progressive needle-in-haystack benchmark showed why a model that loads at 96K context stops staying coherent past 64K—and how we ruled out GPU limits.
Loom Mills: From Agent Swarms to Software Production Lines
How Loom Mills turns roadmap intent into reviewed, gated, observable work—and why production-line control loops fit better than agent-swarm metaphors.
The First 90 Days: Introducing AI-Assisted Dev to a New Team
A 90-day rollout for AI-assisted development: start with bounded habits, measure the work, and standardize only after the team has evidence.