Field notes from operating private AI infrastructure, GPU inference, agent tooling, and healthcare integration systems. For longer implementation write-ups, see the case studies or the writing hub.
30 posts, newest first
revised Sep 12
Professional · 9 min read
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.
ai-agents · mills · loom · fi-fhir
revised Sep 12
Lab · 6 min read
MCP shipped a breaking specification revision. I wrote the gap analysis, filed it as factory backlog, and Mills started rebuilding the SDK it runs on.
mcp · ai-agents · loom · mills
revised Sep 13
Professional · 4 min read
A design pattern for supervised AI investigations: scoped read access, reviewed drafts, and failure tests before expanding permissions.
ai-agents · llm · guardrails · mcp
revised Sep 12
Lab · 16 min read
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.
loom · loom-core · mills · ci
revised Sep 12
Professional · 5 min read
Four defect classes that healthcare interchange parsers can accept silently, with synthetic samples, real linter output, and the pre-send gate that catches them.
healthcare · edi · x12 · validation
revised Sep 12
Lab · 12 min read
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.
flexinfer · prefix-caching · fp8 · speculative-decoding
revised Sep 12
Lab · 13 min read
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.
flexinfer · long-context · multi-agent · prefix-caching
revised Sep 12
Lab · 5 min read
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.
gpu · kubernetes · rocm · sunshine
revised Sep 12
Lab · 10 min read
A July 2026 platform snapshot: multimodal inference, Mills delivery, flight recording, and the operational gaps that remained after the demos worked.
flexinfer · loom · loom-core · mills
revised Sep 12
Lab · 5 min read
Needle-retrieval tests on a 35B model passed near 63,000 prompt tokens and failed at longer contexts, even when vLLM loaded without a memory error.
flexinfer · vllm · long-context · rope
revised Sep 12
Lab · 7 min read
The May 2026 design for Loom Mills: turn roadmap intent into scoped work, verify each change, and keep costs, failures, and operator controls visible.
loom · loom-core · agents · orchestration
revised Sep 12
Professional · 8 min read
A 90-day rollout for AI-assisted development: start with bounded habits, measure the work, and standardize only after the team has evidence.
agents · ai-assisted-dev · adoption · team-practice