How I built Study Agents with FSRS spaced repetition
Multi-agent architecture, ChromaDB RAG, and FSRS spaced repetition in a production AI learning platform.
The problem
As a Software Engineering student at UPC Barcelona, I studied with scattered PDFs and generic AI chats that didn't remember my level or my material. I needed a system that learned from my notes and helped me retain concepts long-term.
The solution
I built Study Agents as the capstone of the 10X AI Developer program at the Artificial Intelligence Institute:
- 8 specialized agents orchestrated with LangChain and FastAPI
- RAG on user PDFs with ChromaDB and embeddings
- Adaptive tests and real-time educational feedback
- Next.js 15 frontend integrated into my portfolio with Google login
Tech stack
| Layer | Technology |
|---|---|
| Frontend | Next.js 15, React, TypeScript |
| Backend | FastAPI, Python 3.11, LangChain |
| AI | OpenAI GPT-3.5/4, ChromaDB |
| Auth | NextAuth.js (Google) |
Results
- Platform in production at paupedrejon.com/study-agents
- Public demo accessible for recruiters and professors
- Scalable architecture ready for new features (FSRS, flashcards, code interpreter)
Next steps
Iterate with spaced repetition (FSRS) algorithms to optimize when to review each concept, turning Study Agents into a tutor that doesn't just explain, but remembers what you forget.