Eigen

Inspiration

It's 2 AM. You have exam is in a few hours. You remember reading something about a concept earlier in your textbook or notes, but now you are scrolling through hundreds of pages trying to find it again. Traditional tools like PDF readers rely on simple keyword searches, which often fail when the wording is different from what you remember.

We built Eigen to solve this problem. Our goal was to create a tool that understands the meaning behind what students are searching for, rather than just matching exact words. By combining semantic search with a unified document viewer, Eigen helps students quickly find the information they need across all their learning materials.

What it does

Eigen is a platform that allows students to upload learning materials such as PDFs, EPUB books, TXT notes, and MP4 lectures and search them using semantic search.

Instead of scanning for exact keywords, Eigen converts documents into vector embeddings and searches based on conceptual similarity. This means students can search for ideas in natural language and retrieve the most relevant passages even if the wording is different.

Key features include:

Semantic search across uploaded study materials
Built-in viewer for PDF, TXT, EPUB, and MP4 files
Highlighting and annotation tools for notes
AI-powered study tools such as summaries and quizzes
Fast retrieval of relevant sections directly inside the viewer
Eigen transforms static study materials into an interactive knowledge system.

How we built it

Eigen is built using a modern AI-powered retrieval architecture.

When a document is uploaded, the system processes it through several steps:

The document is parsed and divided into smaller chunks of text.
Each chunk is converted into a vector embedding representing its semantic meaning.

These embeddings are stored inside ChromaDB, a vector database optimized for similarity search.

When a user performs a search, the query is also converted into an embedding using the Railtracks framework. The framework then finds the closest matching document chunks using vector similarity. Railtracks was integral in allowing for quick and effective search in our platform through our vector databases.

The frontend provides a unified viewer that supports multiple file formats, allowing users to preview and interact with retrieved results directly inside the application.

AI-powered study tools are built using Gemini, which generates summaries, quizzes, and explanations grounded in the user's uploaded content.

Challenges we ran into

One major challenge was handling large document files, especially textbooks that can contain hundreds of pages. To make semantic search efficient, documents had to be split into smaller chunks while maintaining context.

Another challenge was supporting multiple content formats in a single viewer. PDFs, EPUB files, text documents, and video lectures all require different rendering approaches and libraries. Integrating these formats into a seamless reading experience required careful UI and component design.

We also had to ensure that semantic search results could link back to the exact location inside a document, allowing users to immediately see the relevant passage.

Accomplishments that we're proud of

We are proud of building a working semantic search system that retrieves information based on meaning rather than simple keyword matching. Seeing relevant passages returned from large documents was a major milestone.

We also successfully created a unified viewer that supports multiple formats including PDF, TXT, EPUB, and MP4, allowing users to interact with all their study materials in one place.

Another accomplishment was integrating AI-powered study tools that generate summaries, quizzes, and key concepts directly from uploaded content.

Overall, we are proud that we were able to build a complete end-to-end prototype during the hackathon that demonstrates how AI can transform the way students search and study their learning materials.

What we learned

Through building Eigen, we learned a great deal about:

Designing and implementing semantic search systems
Working with vector embeddings and similarity search
Building scalable retrieval pipelines using vector databases
Creating unified viewers for multiple document formats
Integrating AI tools in a way that keeps responses grounded in the user's content

Most importantly, we learned how powerful AI becomes when it is combined with the user's own knowledge base, rather than relying solely on general internet data.