-
My Data Wasn’t Mine. So I Fixed That.
A city CIO’s case for owning your own data — and what I built to prove it works.
-
Data-Driven vs. Data-Informed: Why the Difference Actually Matters
Data-driven means the data makes the call. Data-informed means humans make the call, guided by data. For government and municipal IT, the distinction matters more than you think—and neither works without clean data foundations.
-
Split-Brain AI: When Your Assistant Forgets Itself
TL;DR: I run a self-hosted AI assistant using OpenClaw, accessed via Telegram. After weeks of normal operation, the assistant suddenly started alternating between remembering everything and remembering nothing. Messages would flip-flop—one response coherent and contextual, the next completely blank. Hours of frustrating troubleshooting later, I discovered a MacBook Pro had been turned back on, running a…
-
NVIDIA AI Enterprise Licensing: A Deep Dive for DGX Users
TL;DR: I investigated NVIDIA AI Enterprise (NVAIE) licensing while building local AI infrastructure for a municipal environment. NVAIE costs $4,500/GPU/year or $22,500 perpetual. The DGX Spark has a carve-out that lets you use NIM for free as a single-user workstation, but the line between “workstation” and “server” is vague, and serving multiple users likely triggers…
-
Evaluating GIS AI Models for NVIDIA DGX Spark Platform
TL;DR: Evaluated GIS-specific AI models on DGX Spark for municipal spatial analysis. The hardware performed fine. The model ecosystem did not. Pretrained geospatial models don’t align well with municipal data (wrong resolution, wrong formats, no turnkey pipelines). Deferred GIS workloads; using the hardware for other AI tasks where the ecosystem is more mature. Will revisit…
-
Farm Life AI: What Agriculture Taught Me About Building Useful Systems
TL;DR: I run a farm and work in municipal IT. I tried applying AI tools and thinking to farm problems. Most assumptions from office environments broke immediately. Connectivity, data quality, and hardware durability all fell short. The experiments that worked were simple, reliable, and kept me in the loop. The ones that tried to be…
-
Minutes Machine: Why I Shelved AI-Generated Meeting Minutes
TL;DR: I built a local AI pipeline to draft municipal meeting minutes from recordings. It worked well enough on clean demos but degraded quietly on real meetings: speaker misattribution, merged discussion points, smoothed-over disagreements. The deeper problem wasn’t accuracy, it was accountability. Official minutes carry legal weight, and nobody could clearly own an AI-generated draft.…
-
Conversational Engineering Methodology
TL;DR: I’m developing AI systems for a small city government and kept running into the same problems: models fabricating municipal data, ignoring safety constraints, drifting off track in long conversations. Ad-hoc prompting didn’t scale. What worked was treating conversations like engineering problems—explicit boundaries, structured inputs, adversarial testing, and pipeline-level controls. It helped. It didn’t solve…
-
AI Governance Framework Documentation
TL;DR: I started developing AI for a small city and quickly learned that ad-hoc decision-making doesn’t hold up when CJIS data, public-facing chatbots, and shadow AI are all in play at once. So I started writing things down. Not a policy manual, but working documentation of what the AI systems can and can’t do, who’s…
-
Considerations for City and School IT Collaboration
City and school IT often duplicate infrastructure and security effort. Alignment can reduce risk and pool expertise, but consolidation has real tradeoffs and depends on local context.