Scholia: an AI study partner that answers from course knowledgebase

Description and use

Scholia is an AI study partner that answers questions using only structured course material, providing precise citations to lectures, slides, and textbook sections.

Problem

Students need a study partner that can answer questions using only the specific course material and provide precise citations to lectures, slides, and textbook sections.

Outcome

Scholia delivers precise, source‑cited answers for five study modes, enabling students to study efficiently and understand where to review or improve. It also allows course staff to review external references added by students.

Workflow

• 1. Store course content as typed Sanity documents (lectures, slides, book chapters, assignments, submissions).
• 2. Build a Sanity Knowledge Base that links segments, slides, and book sections to shared topic documents.
• 3. Use GROQ queries to walk from grader feedback to rubric criteria to topics to content locators.
• 4. Run an agent (Vercel AI SDK + Amazon Bedrock) that answers user queries by selecting the appropriate mode and returning citations.
• 5. If a topic is not covered, query a short allowlist of trusted sites via Tavily and present results in an “Outside course material” section.

Evidence

Scholia is a study partner for one course. You ask it something, and it answers from that course's own material: the lecture recordings, the slide decks, the textbook, the problem sets and their rubrics, and your own graded work.

Source

It has five modes, and each one only works because the content is structured: Study, Assignment, Improve, Revise, Mock exam.

Source

When the course does not cover something, Scholia searches a short allowlist of trusted sites (docs.python.org, ocw.mit.edu, Wikipedia and a few universities). Those results never mix with course citations.

Source