Research

A retrieval layer for source-aware courses.

The course engine can use mapped and scraped pages, videos, and domain context so lessons are not isolated model guesses.

01

Source discovery

Queries should be specific to the course and lesson.

Strong courses need diverse, relevant sources instead of repeating the same generic URLs across pages.

  • Topic queries
  • URL dedupe
  • Relevance scoring
02

Content grounding

Sources support claims, not decoration.

Citations should map to the paragraph or concept they support so the learner can trust why material appears.

  • Claim-level evidence
  • Source metadata
  • Graceful low-source states
03

Media retrieval

Videos need a purpose inside the lesson.

A video is useful when it explains, demonstrates, contrasts, or gives a concrete example tied to the lesson objective.

  • Relevance reason
  • Duration fit
  • Transcript-aware summaries