Skip to content

Add U3 Pro voice agent recommendation to model selection page#783

Open
agalyanmann wants to merge 1 commit intomainfrom
bearclaw/docs-fix/model-selection-voice-agent-recommendation
Open

Add U3 Pro voice agent recommendation to model selection page#783
agalyanmann wants to merge 1 commit intomainfrom
bearclaw/docs-fix/model-selection-voice-agent-recommendation

Conversation

@agalyanmann
Copy link
Contributor

Summary

  • Adds a <Tip> callout after the model comparison table on the model selection page (select-the-speech-model.mdx) explicitly recommending Universal-3 Pro Streaming for voice agent use cases.
  • The page previously listed models neutrally with no guidance on which to choose for voice agents. Evaluators defaulted to Universal-Streaming English and got poor results on short utterances and entity recognition.

Evidence

Multiple voice agent evaluators used the default model (Universal-Streaming English) instead of U3 Pro, resulting in poor out-of-box accuracy. Internal confirmation: "Default should be U3 Pro with proper prompts" and "for short utterances, that default prompt is going to work way better than anything else a customer tries."

Test plan

  • Verify the <Tip> renders correctly on the docs site
  • Verify the link to /docs/voice-agent-best-practices resolves

🤖 Generated with Claude Code

The model selection page lists models neutrally without guiding voice
agent builders toward U3 Pro. This adds a tip after the comparison
table explicitly recommending U3 Pro for voice agents, with a link
to the best practices guide.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
@github-actions
Copy link

agalyanmann pushed a commit that referenced this pull request Mar 18, 2026
Route healthcare/medical scribe builders to the Medical Scribe Best
Practices guide from the three pages they are most likely to land on:
the async getting-started quickstart, the models overview (Universal-3
Pro section), and the pre-recorded evals guide. Also adds a section on
domain-specific evaluation beyond WER, since medical scribe customers
often evaluate based on user preference rate and readability rather
than word error rate alone.

Follows the same <Tip> callout pattern used for voice agent routing
in PRs #782 and #783.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant