Blog
Neo: an always-on engineering copilot on our own GPUs
How a LangGraph agent with local models, MCP tools, and an ideas loop became the rig's resident engineer — and why owning the inference changes what a copilot can be.
Shellback: the ops shipmate that never sleeps
Why we gave the GPU rig a Navy-grade watchstander — an MCP-native ops agent that answers every AI client with the same tools, memory, and no-nonsense attitude.
SlopeyOps: an iPhone ops console for a one-man engineering shop
Building a personal iOS app in an afternoon — Neo chat, GPU monitoring, site change-notes, and a photo pipeline for blog assets — with graceful offline states because the rig was literally down while we built it.
Building ENCORE — a Rock-Stage Piano Rhythm Game for iOS
How I designed and built ENCORE, a touch-tuned piano rhythm game, end-to-end — and shipped it to TestFlight.
Self-Hosting Open-Weight LLMs
When running Qwen/Llama on your own GPUs beats a frontier API — cost, privacy, quantization tradeoffs, and routing between local lanes and hosted models.
CI/CD for Small Shops
A pragmatic pipeline: trunk-based dev, fast tests, preview deploys, and one-command rollback — what to automate first when you don't have a platform team.
Motor Control Lessons From the Bench
Practical STM32/ESP32 notes on PWM, current sensing, debouncing, and brownouts — and why the datasheet is never the whole story.
One Beefy Box vs. Kubernetes
The case against over-engineering: when a single well-provisioned server beats a cluster, and the signals that say it's actually time to scale out.
Agentic Systems That Don't Go Rogue
Building AI agents that propose actions but require human approval before doing anything irreversible — approval gates, dry-run demos, and audit trails.
From Firmware to Cloud: Full-Stack Includes the Hardware
Taking a product from STM32/ESP32 firmware and custom PCBs up through edge telemetry to a cloud dashboard — and why owning the whole stack removes finger-pointing.
What SDVOSB Status Actually Means for a Federal Contract
Set-asides, the trust a veteran-built shop brings, and why agencies choose Service-Disabled Veteran-Owned vendors.
On-Prem RAG for Regulated and Federal Workloads
Why retrieval-augmented generation on hardware you control beats shipping sensitive documents to a SaaS API.
One MCP Server, Every AI Client
How we wrapped GPU status, image generation, and memory into one Model Context Protocol server so Claude, Cursor, and our own agents call the same tools.
Why We Run Our Own GPU Rig
The economics, latency, and data-control case for on-prem inference at a small shop.