AgoraLearn

What will inspire me / what inspires this build I see classmates lose time and marks on lab graphs—scales, units, graph-paper-style layouts, and best-fit / regression lines—while still not getting a clear picture of the experiment. Flat textbook diagrams also make optics, projectile motion, and molecular geometry harder to internalize. This project will be my answer: a single workflow where students upload data or notes, ask in text or voice, and receive interactive graphs and 3D simulations that match how science is taught and assessed.

What I will learn I will learn to implement a retrieval-style (RAG) pipeline end to end: chunking and embedding uploaded material, retrieving the right context, and generating answers that stay grounded in the student’s documents. I will learn to design structured outputs (e.g. JSON for charts and 3D scenes) so the UI can render results reliably. I will also learn to integrate voice, multilingual translation, and robust file ingestion without making the user experience feel fragmented.

How I will build it I will use a Next.js frontend for chat, uploads, interactive charts, a 3D viewer, and voice controls. I will connect it to a Node/Express-style backend with APIs for ingest, query routing, and voice. I will store chunked text and vector embeddings in Supabase for similarity search. Gemini will be the primary model for reasoning, multimodal understanding, and structured generation; Lingo will handle translation across languages. I will define clear request/response contracts for each mode (Q&A, plotting, 3D, tables, voice).

Challenges I faced I struggled early with keeping outputs consistent—chart and 3D payloads had to validate and render every time. Messy real-world uploads (PDFs, scans, pasted tables) forced me to rethink extraction and error messages. Voice end-to-end introduced latency, failure cases, and UX edge cases I did not anticipate at first. I also hit friction with routing the right “mode” (Q&A vs graph vs simulation) from informal student phrasing, and with deployment basics (environment variables, CORS, upload limits) until I tightened the architecture.

Built With

• chart.js
• next.js
• node.js
• react
• supabase
• typescript