🎓

UNIVY

AI-Powered Study Helper

Transform lecture scripts, slides, and altklausuren into structured study materials

🎯 Vision

To help students learn smarter by automatically generating traceable, structured, and context-aware study materials from their university resources. By combining document parsing, semantic analysis, and knowledge graph building, UNIVY bridges the gap between passive reading and active learning.

📌 What It Does
1

Uploads and parses scripts, slides, altklausuren into structured document representations.

2

Generates semantic notes with clickable references to the original source.

3

Creates flashcards from lecture content and old exam questions.

4

Answers altklausur questions based on provided scripts and slides using document-grounded RAG.

5

Builds a knowledge graph to visualize key concepts and their relationships.

✅ Core Features

Docling PDF Viewer

View scripts, slides, and exams with clickable highlights and anchors

Smart Notes

AI-generated, context-aware notes linked to document locations

Flashcard Generator

Creates Q&A pairs from parsed material for active recall

Altklausur Answering Mode

Automatically answers past exam questions using LightRAG

Node-Entity Graph

Visual map of entities and their connections from uploaded content

Source Tracing

Notes and flashcards are always linked back to script/slides

🧱 Tech Stack

🖥 Frontend: React

  • • Viewer: Docling-based lightweight PDF viewer
  • • UI: MUI / Tailwind (customizable)
  • • State Management: Zustand or Context API
  • • Components: Note Panel, Flashcard Deck, Graph View, Altklausur Answerer

⚙️ Backend: FastAPI

  • • Upload & Document Ingestion: FastAPI routes for file parsing
  • • RAG Engine: LightRAG for document-grounded Q&A and flashcard support
  • • Worker Queue: Celery

🧠 AI & Embeddings

  • • Document Parsing: Custom logic + LLM + PDFMiner/MinerU
  • • Sentence Chunking: spaCy or LLM tokenizer
  • • Embedding: OpenAI Embeddings / LightRAG
  • • QA / Notes / Flashcards: GPT-4-turbo / Claude / Custom Prompts
  • • Node-Entity Graph: Entity linking via LLM + Neo4j or simple JSON graph schema
🔁 System Flow
  1. 1.User uploads PDF → FastAPI receives & stores document.
  2. 2.Document is parsed → Sections, headings, and sentences extracted.
  3. 3.Content is chunked & embedded via LightRAG.
  4. 4.Notes and flashcards are generated from embeddings and semantic cues.
  5. 5.Altklausur mode extracts questions, answers using grounded QA.
  6. 6.Entity graph builder maps key terms and relationships.
  7. 7.Frontend renders all outputs, traceable and interactive.
📚 Example Use Case

Upload your ML lecture script, slides, and altklausur. UNIVY parses everything and generates:

  • • A sidebar with highlighted smart notes linked to the PDF
  • • A flashcard deck based on key concepts
  • • Altklausur answers grounded in script context
  • • A graph view showing key topic relationships