Resources

Essays, product notes, and operating lessons from the build.

We write about architecture, voice systems, deal intelligence, product design, and the tradeoffs behind each release.

At a glance

Published9

Latest entryMay 23, 2026

Recent themes

AnthropicEvalsDental AIProduction

May 23, 2026Bryan Mathews

199 Evals Before Our Dental AI Answers a Call

Velyn Dental runs 22 eval suites and 199 cases on every change to its Claude runtime. This post walks through what the suites cover, how the redaction policy lets us commit results, and why offline coverage beats a live demo.

AnthropicEvalsDental AIProduction

May 23, 2026Bryan Mathews

How We Built a Claude-Native Dental AI Receptionist

The architecture decisions behind Velyn Dental's Claude-native runtime — three planes (runtime, MCP, GitHub), the migration off mixed-provider speech, and the deterministic safety floor under the model.

AnthropicClaudeArchitectureDental AI

May 23, 2026Bryan Mathews

The Five Governance Exceptions in Our Dental AI Receptionist

Velyn Dental routes most AI output through human review checkpoints. Five workflows are documented exceptions — they fall back deterministically instead. This post explains each one, the deterministic safety floor behind it, and the overclaim guardrail.

AnthropicGovernanceCOEDental AI

October 22, 2025Bryan Mathews