Prepfolio
AI adaptive learning platform with RAG system, multi-model AI architecture, and ELO-based personalized learning.
What I did
Architected full-stack AI-powered adaptive learning platform using Next.js 15, React 19, TypeScript, Tailwind CSS v4, and Supabase
Designed and implemented production-grade RAG system with 6 specialized chunking strategies, achieving <200ms vector search latency using Qdrant
Built multi-model AI architecture integrating OpenAI GPT-5 Mini, Anthropic Claude Haiku 4.5, and Google Gemini with streaming SSE responses
Implemented semantic chunking algorithm using OpenAI embeddings (1536 dimensions) with cosine similarity-based boundary detection
Developed ELO-based adaptive learning algorithm (0-1500 scale) with 7 performance tiers and real-time ability tracking
Created prompt caching infrastructure with memory and Redis backends, reducing AI API costs by ~40%
Developed Duolingo-style learning path visualization with winding chapter maps and streak-based progress tracking
Built spaced repetition system using SM-2 algorithm for flashcard scheduling with dynamic ease factor adjustment
Implemented multi-format document viewers (PDF, DOCX, PPTX) with text selection, annotation support, and citation highlighting
Built 60+ reusable React components with shadcn/ui primitives