Blog
ALL POSTS
Essays and articles on AI, data, and the infrastructure underneath.
AI: From the Engine Room
A DATA LEADER'S PERSPECTIVE ON MODERN AI
AI Mechanisms
I've Heard This Engine Before
I worked on attention mechanisms before ChatGPT made them famous. That perspective has shaped how I separate AI signal from noise.
The Attention Mechanism Explained
"Attention" is a brilliant name for what these models do, and also misleading in ways that matter.
Training Economics
"Let's just train a model on our data" sounds straightforward. The actual cost structure tells a different story.
The Efficiency Illusion
That benchmark showing the smaller model matches GPT-4? Probably accurate. Also hiding tradeoffs that only surface in production.
Context Windows and Memory
The most common misconception I encounter: people assume these models remember previous conversations. They don't.
The Proprietary Data Paradox
The Multiplier Effect
Proprietary data isn't the AI advantage most organizations think it is. What that data connects to is.
The Tacit Knowledge Bottleneck
The most valuable knowledge in any organization is often the hardest to capture, because the people who have it can't fully explain it.
Why Vectors Aren't Enough
Vector search finds what's similar. Expertise often requires connecting things that aren't similar at all.
UI as the Ultimate Guardrail
AI systems speak with uniform confidence whether they're right or wrong. Interface design determines whether that difference becomes visible.
Forward-Looking Governance
Hallucination is a Feature, Not a Bug
Vendors frame hallucination as a bug they're fixing. It's actually inherent to how these systems work, which changes the governance question entirely.
Working With the Machine
The most valuable prompting skill isn't writing better prompts. It's knowing when prompting isn't the answer.
AI Readiness Is a Governance Question
Every AI conversation eventually becomes a data conversation. Every data conversation eventually becomes a governance conversation. The sequence matters.
What the Engine Room Taught Me
These systems are more comprehensible than marketing suggests, and more limited than hype implies. Both are true simultaneously.
Going Emic
LEARNING A TECHNICAL DOMAIN THROUGH ITS OWN CATEGORIES
The Insider's View
Learning a Domain From Its Own Categories
The hardest part of cross-industry research is learning a new field fast. There are two ways in, and one of them is built from the domain's own words.
Why I Picked the Most Impenetrable Ontology I Could Find
If a comprehension tool can make DMTF's Common Information Model legible, everything easier comes for free. So that's where I started.
Building the Explorer
Shipping a Triplestore App With No Server
On paper Emic is a database application with four moving parts. In production it's a folder of static files. Here's how the backend got compiled away.
Teaching a Graph to Rank Itself
A thousand-node ontology is a hairball until you ask it the right questions. Three graph algorithms and one design lesson turn it into a map.
The Org Chart vs. the Kinship
Every ontology declares how it's organized. Its concepts have an actual shape that's often different. You can measure the gap, and CIM's is wide.
Field Notes
The $95 Education of an LLM Cache
Pre-writing 6,000 field notes with an LLM cost more than the plan said and taught more than expected, including which model quietly narrates its own prompt.
What an Ontology Explorer Taught Me
I set out to build a tool that explains ontologies. I ended up building one that's careful about how it explains them, and that was the harder, better project.