A practical FinOps playbook for Databricks. Cluster types, file compaction, caching, serverless, and BI rationalization — with realistic savings ranges.
Air-Gapped Agentic Stacks for Sovereign Environments
Sovereign-AI deployment is now operationally feasible. Here is what an air-gapped agentic stack looks like, what it costs, and where it fits in federal mission environments.
Power Bi Direct Lake on Databricks SQL: a Modernization Playbook
Migrate Power BI semantic models from import / DirectQuery to Direct Lake on Databricks SQL. Performance, governance, and migration patterns.
Calibration-First AI for Federal Decision Support
Federal decision support cannot run on headline accuracy. Calibration and conformal prediction are the standards a procurement officer should require — and the standards we hold ourselves to.
What We Open-Sourced This Year — and Why
Four open-source projects, four different reasons. A short manifesto on what we open-source, what we don’t, and why.
Production ML on Databricks: Mlflow, Feature Store, Calibration
A reference MLOps stack on Databricks — MLflow Model Registry, Feature Store with online serving, calibration-first model evaluation, and Mosaic AI Model Serving.
Building Multi-Agent Workflows on Databricks (mosaic AI Agent Framework)
Multi-agent workflows native to the Lakehouse — designed, built, evaluated, and deployed on the Mosaic AI Agent Framework with typed tools and an evaluation harness.
When Agents Call Agents: Why the MCP Server Matters in Freight
Model Context Protocol lets external AI agents call FreightCortex tools natively. Here is why that matters — and what it unlocks for the freight intelligence stack.
Hybrid Retrieval: Why Vector Alone Isn’t Enough
Vector search is excellent at semantic similarity and bad at named entities. BM25 is the opposite. Production-grade retrieval is hybrid — and the architecture decisions matter.
Why Calibration Matters More Than Accuracy: an ECE 0.012 Story
Headline accuracy is a misleading metric for high-stakes decisions. Calibration is the real one. Here is what ECE 0.012 means and how we got there.