Smart Study Buddy

Inspiration

The "Information Age" has left students with a common problem: Information Overload. We noticed that students spend 80% of their time organizing notes and only 20% actually learning them. We wanted to flip that ratio. Inspired by the power of open-source AI, we built Smart Study Buddy to act as an intelligent filter—turning overwhelming PDFs into structured, actionable knowledge.

What it does

Smart Study Buddy is an AI-driven learning ecosystem that transforms static documents into interactive experiences.

Instant Synthesis: Upload any PDF or text file to get a structured summary.

Knowledge Extraction: Automatically generates Q&A pairs to test user comprehension.

Contextual Chat: A built-in tutor that answers specific questions based only on the uploaded material.

On-the-Go Learning: Key concepts are distilled into high-impact bullet points for quick review before exams.

How we built it

We prioritized a Python-native stack to ensure speed and seamless integration:

Frontend & UI: Built with Streamlit, allowing for a highly interactive, responsive dashboard with minimal overhead.

AI Models: We leveraged the Hugging Face ecosystem, using state-of-the-art Large Language Models (LLMs) for summarization and text generation.

Text Processing: Utilized LangChain to handle document loading and recursive character splitting.

Inference: Connected to the Hugging Face Inference API to run heavy-duty models efficiently in the cloud, ensuring a smooth experience even on low-end hardware.

Challenges we ran into

The most significant hurdle was Token Management. Hugging Face models have specific "context windows"—trying to feed an entire 50-page textbook into a model at once would cause it to fail. We overcame this by implementing a Recursive Text Splitting logic that breaks down the text into overlapping chunks, ensuring that no context is lost at the "seams" of the data. We also spent significant time fine-tuning the Temperature and Top-P parameters of our models to ensure the summaries were factual and not "hallucinated."

Accomplishments that we're proud of

Latency Optimization: We successfully implemented caching mechanisms in Streamlit so that once a document is processed, switching between summaries and quizzes is instantaneous.

Clean UX: Moving from a raw Python script to a polished, intuitive web interface that anyone can use.

Zero-Shot Accuracy: Achieving high-quality quiz generation without needing to train a custom model, purely through clever prompt engineering and the power of Hugging Face’s open-source models.

What we learned

We gained deep experience in the RAG (Retrieval-Augmented Generation) workflow. We learned that the quality of AI output is directly tied to the quality of the data "chunking" and cleaning. We also discovered the incredible speed of the Streamlit development cycle, which allowed us to iterate on our UI based on real-time testing during the hackathon.