I Built a Local AI Study Organizer for My Friend's Messy PDF Library

MANAV mishra — Sun, 04 Oct 2026 16:25:20 +0000

This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend.

What I Built

A friend of mine has a large collection of study PDFs spread across different folders. The problem wasn't a lack of study material. The problem was finding the right material quickly and knowing what each document actually contained.

So I built StudyShelf AI, a local-first study organizer that uses an open-weight AI model to turn a messy collection of study PDFs into an organized, searchable study shelf.

The idea is simple:

Instead of manually opening dozens of PDFs to figure out what they contain, upload them to StudyShelf AI and let local AI organize them.

For each document, the application generates:

Subject/category
Relevant tags
Priority
Short summary

The result is a much easier way to understand and search through a large study library.

Demo

Video demo:

Watch the StudyShelf AI demo

The demo shows the complete workflow:

Running the Qwen model locally through LM Studio
Opening the StudyShelf AI application
Uploading a study PDF
Extracting the PDF text locally
Sending the extracted text to the local AI model
Generating the category, tags, priority and summary
Adding the document to the searchable study shelf

Code

GitHub repository:

MANAV-MISHRA-BYTES/StudyShelf-AI
(https://github.com/MANAV-MISHRA-BYTES/StudyShelf-AI)

The complete source code, requirements, setup instructions and project documentation are available in the repository.

How I Built It

StudyShelf AI is built with Python and Streamlit.

The core pipeline is:

Study PDF
   ↓
PyMuPDF
   ↓
Local text extraction
   ↓
LM Studio
   ↓
Qwen3.8 9B
   ↓
Category + Tags + Priority + Summary
   ↓
Searchable Study Shelf
   ↓
CSV Export

Tech Stack

Python
Streamlit
PyMuPDF
Pandas
Requests
LM Studio
Qwen3.8 9B Q5_K_M

PyMuPDF handles local text extraction from PDF files.

Streamlit provides the user interface.

Requests communicates with the local LM Studio API.

Pandas is used to organize and export the generated document index.

LM Studio runs the open-weight model locally and provides the local API used by the application.

For this project, I used Qwen3.8 9B Q5_K_M through LM Studio.

Why Does Open Innovation Matter?

Open-weight AI was important to this project because the application deals with study documents that may contain private notes, assignments, personal annotations and other information that a student may not want to upload to a third-party AI service.

With StudyShelf AI, the core document organization workflow can remain on the user's computer:

Study PDF
   ↓
Local text extraction
   ↓
Local LM Studio server
   ↓
Open-weight Qwen model
   ↓
Document organization

There is no requirement for a cloud AI API for the core organization task.

This gives the user more control over their study material and avoids sending the documents to an external AI service just to categorize and summarize them.

It also means the model can be changed or upgraded without rebuilding the entire application around a proprietary API.

For this project, open AI was not just a requirement of the challenge. Local inference directly supported the privacy goal of the application.

What I Learned

One of the biggest lessons from this project was that a useful AI application does not necessarily need to be huge.

I initially considered building a much larger study assistant with RAG, a database, authentication, question answering and several other features.

But that would have added complexity without solving the immediate problem better.

Instead, I focused on one useful workflow:

Take a messy collection of study PDFs and make it understandable.

That kept the application small enough to build during the challenge while keeping the AI component central to the solution.

Current Limitations

StudyShelf AI is currently an MVP, so there are several things it does not do yet:

Scanned or image-only PDFs are not currently OCR'd.
It does not yet provide full RAG-based question answering.
The organized library is not currently persisted in a database.
The application expects a local LM Studio-compatible server.
Search currently operates on generated document metadata rather than performing semantic search across the full document contents.

These limitations also provide a clear path for future versions.

What's Next?

Some improvements I would like to add in future versions are:

OCR support for scanned documents
Persistent document libraries
Semantic search
RAG-based question answering
Flashcard generation
Quiz generation
Automatic revision scheduling
Duplicate document detection
Better document previews
Support for additional local open-weight models

Hacktoberfest

StudyShelf AI was built for the Hacktoberfest Weekend Challenge: Build for a Friend.

The project addresses a real study-organization problem for a friend while using an open-weight AI model and local inference at the core of the solution.

The challenge encouraged building something useful for a real person rather than simply building a large project for the sake of complexity.

I therefore focused on one small but practical problem: making a large collection of study material easier to understand and navigate.

Prize Categories

No partner category is being claimed for this submission.

My Agent Session

No agent session is being submitted.

Thanks for checking out StudyShelf AI.

DEV Community: MANAV mishra