LumiAI

• 

  Main page

• 

  After uploading PDF

Inspiration

University students and self-learners routinely face cognitive overload when tackling 50+ page course slide decks, especially when studying technical subjects across multiple languages. Synthesizing essential concepts, isolating terminology, and hand-crafting revision flashcards takes hours away from actual learning. We built Lumi to streamline that entire cycle into an interactive, multilingual study workflow.

What it does

Lumi is a focused study assistant built with Streamlit that processes course PDFs and extracts three structured study assets:

Concise Summaries: High-yield breakdowns highlighting core exam concepts without fluff.

Interactive Flashcards: Question-and-answer pairs with hidden answers for active recall testing.

Key Definitions: Isolated technical vocabulary for rapid reference and revision.

Multilingual & Translation Support: Full support for generating and translating materials across English, French, and Arabic.

Offline Study Packs: One-click downloads to save study decks as local text files for offline review.

How we built it

PDF Extraction: Local document parsing to ingest university slides and extract clean text streams without external cloud converters.

LLM Orchestration: Powered by Google Gemini to parse content, enforce strict structural schemas for interactive components, and handle cross-lingual translation.

Frontend Architecture: Streamlit interface styled with a focused dark mode, progress indicators, interactive reveal states, and export mechanisms.

Challenges we ran into

Preserving structured data formatting across three distinct languages (particularly Arabic layout and translation nuances) while ensuring flashcard states remained interactive was tricky. We solved this by strictly validating the data schemas between the extraction pipeline and the Streamlit UI components.

What's next for Lumi

Anki deck (.apkg) export compatibility.

Multi-file batch processing for full semester slide archives.

Local offline small-LLM inference support.

Built With

• education
• gemini-api
• multilingual
• pymupdf
• python
• streamlit