Prism - Study Workspace

Prism turns any course material into a personalised learning journey, complete with source-grounded AI help, adaptive quizzes, flashcards, notes, and on-demand whiteboard video explanations.

Comment

Inspiration

Students often have plenty of material but no clear path through it. Prism helps turn scattered notes, links, and course content into focused, actionable learning.

What it does

Prism organizes courses and resources, then creates personalized learning journeys with lessons, milestones, quizzes, flashcards and video explanations. Students can ask source-grounded questions and request whiteboard-style video explanations for difficult concepts.

How we built it

We built Prism with Next.js, React, TypeScript, and Cloudflare. D1 and Drizzle manage study data; AI Search/AutoRAG grounds answers in student materials; and Queues and Workflows power ingestion and long-running AI tasks. OpenAI generates structured study content, and creates whiteboard videos.

Codex CLI was a key development partner: its repo-aware code generation, multi-file edits, terminal- based testing, planning, debugging, and code-review workflows helped me move quickly across the full stack. Its subagent workflow support was especially useful for parallel codebase exploration and reviewing independent parts of the system.

Challenges we ran into

Reliable AI output was our biggest challenge. I needed study plans, quizzes, and flashcards to be structured and useful, so we added schemas and validation. We also had to coordinate asynchronous indexing, content generation, and video rendering while giving users clear job states.

Accomplishments that we're proud of

I built more than a chatbot: Prism connects a student’s own material to a complete learning loop plan, learn, practise, identify gaps, and get an explanation. I am especially proud of the source-grounded AI workflow and resilient background-job architecture.

What we learned

I learned that useful AI products need dependable systems around the model: strong data contracts, retrieval grounding, validation, and transparent loading and failure states. I also learned how much faster product iteration becomes when Codex can help plan, implement, test, and review complex changes.

What's next for Prism - Study Workspace

Next, I want to deepen personalisation with spaced-repetition scheduling, richer progress analytics, collaborative study spaces, and learning plans that adapt continuously to quiz performance and study habits.

Built With

Share this project:

Updates