UniMind — AI Learning & RAG Platform

RAG-based learning platform that transforms lecture documents into grounded Q&A, summaries, quizzes, mind maps, and personalized learning resources.

RAGFastAPIOpenAIGeminiPGVectorEmbeddingsNestJSPostgreSQL

Engineering Decisions

Co-developed a three-person AI learning platform combining NestJS application services with a dedicated FastAPI AI microservice.

Implemented token-aware document chunking targeting roughly 400–500 tokens with overlap for more precise retrieval.

Used OpenAI text-embedding-3-small embeddings with semantic vector retrieval for grounded learning workflows.

Designed structured RAG responses containing answers, confidence levels, and evidence references.

Supported OpenAI and Gemini providers across AI workflows including RAG, summaries, quizzes, and learning-content generation.

Architecture

Lecture PDFDocument ProcessingToken ChunkingEmbeddingsPGVectorSemantic RetrievalLLM

Data & AI Flow

Lecture content is extracted, chunked and embedded, retrieved semantically from the vector store, then passed as evidence to the selected LLM for structured grounded generation.

Engineering Tradeoff

Separated the AI workload into FastAPI rather than coupling generation logic directly to the NestJS application layer.

Proof Points

ArchitectureData FlowTradeoffScale