From 10X course to live product
Turned the Instituto de IA capstone into a deployed, maintained app — not a mockup: production code, video demo and a direct link from projects.
Back to projects
StudyAgents
AI-Powered Learning Assistant
Capstone of the Developer 10X with AI program (Instituto de Inteligencia Artificial): 8 intensive weeks + final project. Multi-agent platform in production inside my portfolio — real study flows, not long chat paragraphs.
Training + capstone
Study Agents is the final project of the Developer 10X with Artificial Intelligence course at the Instituto de Inteligencia Artificial — an 8-week program plus capstone where I consolidated agents, RAG, Python APIs and a Next.js product. I built on what we learned in class (LLM orchestration, pipelines, best practices) and went further on my own: full multi-agent architecture, study-first UX and a real deployment linked from this portfolio. It is my flagship applied-AI piece solving a problem I lived as a student.
The product
The goal: turn AI into an interactive study companion. Users upload PDFs, content is indexed with RAG (ChromaDB) and eight specialized agents cover the full loop — explain at the right level (0–10), answer with context, generate tests and exercises, correct with educational feedback and export notes. FastAPI + LangChain on the backend, Next.js on the frontend, Google login and a public demo recruiters can try in one click.
What I built
8 orchestrated agents
Content Processor, Explanation, Q&A, Test Generator, Feedback, Exercise Generator/Corrector and Correction Agent, shared memory and smart GPT-3.5 / GPT-4 routing.
RAG on your notes
PDFs → chunking → embeddings → ChromaDB. Answers ground on the user's material, not generic hallucinations.
Complete study flows
Chat, Markdown→PDF notes, adaptive tests, language flashcards, in-browser code runner and progress dashboard.
In production
paupedrejon.com/study-agents · NextAuth (Google) · per-user API keys · scalable FastAPI backend.
Portfolio integration
Same Next.js 15 monorepo, i18n, consistent branding and visible CTAs on home and projects.
Tech stack
Clear split: Python orchestrates AI; React/Next.js delivers the experience.
- Next.js 15App Router · UI · deploy
- React & TypeScriptInteractive frontend
- FastAPIAPI & agent orchestration
- LangChain + OpenAILLM pipelines
- ChromaDBVector store · RAG
- NextAuth.jsGoogle sign-in
- Python 3.11Backend & agents
- jsPDF / MarkdownNotes export
My role
Sole capstone owner: multi-agent architecture, Python backend, Next.js frontend, RAG, portfolio integration, deployment, documentation (README + demo) and UX iteration based on real student use.
Case study: Study Agents — AI learning platform
Study Agents transforms PDFs and notes into a personal tutor with eight specialized agents, ChromaDB RAG and complete study flows: chat, tests, exercises and note export.
The problem
Students use generic AI chats that don't know their material, don't adapt explanations to their level (0-10) and don't offer a structured study cycle with tests, correction and progress tracking. As a UPC student, I experienced this daily.
Technical solution
Multi-agent architecture with FastAPI + LangChain: Content Processor indexes PDFs in ChromaDB, Explanation Agent adapts to user level, Q&A answers with RAG context, Test/Exercise Generators create personalized assessments, and Feedback/Correction Agents provide educational feedback. Next.js 15 frontend with NextAuth (Google), per-user API keys and public demo integrated in the portfolio.
Results
Platform in production at paupedrejon.com/study-agents, accessible with one click from the portfolio. Capstone of the 10X AI Developer program at the Artificial Intelligence Institute. Demonstrable for recruiters: not a mockup, maintained code with scalable backend.
- AI agents8 specialized
- BackendFastAPI + LangChain
- Vector DBChromaDB (RAG)
- StatusIn production