Inspiration

A student can watch a lesson and still struggle to understand why their code behaves differently from what they expected. Meanwhile, instructors see grades and scattered questions but have to piece together where the class is struggling. Our experience with CS 1301 inspired Trellis: a course environment that connects explanation, practice, and feedback.

What it does

Trellis is a customizable course builder and learning platform, with CS 1301 as our proof of concept. Instructors organize lessons, videos, questions, coding exercises, and quizzes through the browser. Students have one place to learn, practice, receive feedback, and track progress.

Three features connect that experience:

  • See the program work. An interactive Python visualizer steps through execution, variables, and data structures. Students can edit examples, and instructors can create visualization sandboxes. After correctly completing eligible coding exercises, students can explore a sample solution visually.
  • Find help in context. The embedded discussion board supports shared answers, endorsements, and followups. While students draft a question, related answered posts appear to help them find existing explanations.
  • Turn questions into teaching feedback. Instructors can review groups of related questions, prioritized pain points, and the original discussions to identify concepts that need another explanation.

Trellis complements existing campus systems rather than requiring Georgia Tech to replace Canvas.

How we built it

We used React and TypeScript for the frontend, Django for the backend, and PostgreSQL for course content and learning records. Redis and Celery coordinate background work. Student Python runs in separate execution containers with resource and time limits, while Docker provides a reproducible local environment.

Discussion matching combines local text embeddings, cosine similarity through pgvector, and keyword signals. It does not require sending student questions to an external language-model service. Course-scoped permissions separate student, TA, and instructor access, and a REST API supports course maintenance and reporting.

Challenges we ran into

Connecting the tools was harder than building individual screens. Navigation, saved work, grading feedback, and discussion context all need to stay consistent.

We also had to distinguish questions about the same concept from posts that merely share common words. Linking pain-point groups back to their original posts lets instructors check the evidence instead of relying on an unexplained summary.

Accomplishments that we're proud of

We built a working course experience with a populated CS 1301 proof of concept, browser-based authoring, real Python execution, interactive visualizations, and an integrated discussion workflow.

The platform supports multiple courses, allowing another instructor to build content without changing the platform’s source code. Our discussion demonstration uses clearly labeled fictional posts and participants.

What we learned

Useful educational technology needs to help both sides of the classroom. Students need clearer explanations and actionable feedback; instructors need visibility into recurring confusion and the flexibility to respond.

CS 1301 gives us a concrete starting point. The course builder, native questions, and discussion tools can support other subjects, while Python execution and visualization provide our first subject-specific tools.

What's next for Trellis

Our next step is a faculty-led pilot to evaluate accessibility, performance, student experience, and staff authoring effort. Campus adoption would include approved GT authentication, institutional Canvas authorization, and privacy, retention, and security review.

The build includes a Canvas assignment-publishing preview, providing a path for instructors to author in Trellis while continuing to use campus systems.

Institution-managed hosting could reduce dependence on multiple per-student licenses while allowing course-specific customization. We would validate potential savings against the full cost of hosting, code execution, databases, storage, and maintenance during the pilot.

Share this project:

Updates

Submission history