Best Practices & Tips

Practical guidance for teams running Sudar in production. Pair these notes with animated walkthroughs and the help center.

Course design (Studio)

Start with a clean source

Better inputs produce better outlines.

  • Use a well-structured PDF or article URL rather than a loose paste when possible.
  • Review the generated outline before publishing; adjust module order to match your compliance story.
  • Add one quiz per module so struggle signals feed the learner profile.

Templates and media search

Keep visual consistency without a design team.

  • Pick a persona/template early and stay on it for the path.
  • Use in-editor media search (Pexels, Unsplash, etc.) for royalty-safe assets.
  • Enable the Studio chime on long AI generations so creators know when a draft is ready.

Delivery (Learn)

Let learners switch modalities

Same module, different channel.

  • Tell learners they can use Listen on mobile and Read at a desk.
  • Encourage Cards after a dense Read module for retrieval practice.
  • Point people to the tutor for “explain this” on selected text.

Build tutor memory deliberately

Memory improves when learners interact over time.

  • Use follow-up questions in the tutor so session context accumulates.
  • Set digest cadence on Memory if your org allows LLM profile updates.
  • Respect org policy: some tenants require consent before overlays or inference.

Paths, compliance, and ALP

Paths and due dates

Make mandatory training measurable.

  • Separate mandatory and optional courses on a path.
  • Schedule the compliance-reminders cron with CRON_SECRET and Resend configured.
  • Use analytics drop-off views to fix modules where everyone stalls.

ALP integration hygiene

Stable identity and batched events.

  • Map LMS user IDs to Sudar profiles.id once and keep the mapping stable.
  • Rotate ALP keys from Studio Integrations if a connector is compromised.
  • Batch events at session end where the LMS allows it.

Operations and trust

Production checklist

Studio, Learn, Intelligence, and DNS.

  • Apply database migrations before first learner login.
  • Set SUDAR_INTELLIGENCE_URL and INTELLIGENCE_SERVICE_SECRET on both Vercel projects.
  • Read docs/trust before procurement reviews; link Governance in Studio.

MCP and external AI clients

Creator tools still obey your tenant auth.

  • Use OAuth MCP for ChatGPT; use short-lived tokens for local Cursor configs.
  • Audit MCP tool calls via /api/mcp/audit when you need operator trails.
  • Do not paste service role keys into MCP client env files.