Collaboration with Bill Inmon on Analytical AI—Read Article

BLOG

February 16, 2026·Johnson Kuan

In Tokens We Trust: Field-Level Confidence for LLM Structured Outputs

Every LLM token comes with a confidence signal. By aligning these signals to individual JSON fields instead of averaging across the whole output, we get far more reliable confidence scores for structured outputs.

Read more →
February 14, 2026·Johnson Kuan

From Transactional AI to Analytical AI

Most AI inference today is transactional: one prompt, one response. But enterprises need to process millions of records with AI. That's an analytical workload, and it needs a completely different architecture.

Read more →