Recallify

Inspiration

As students, we often struggle with staying engaged while reviewing static notes. Active recall is one of the most effective study techniques, but it's hard to simulate without a real study partner. We wanted to create something that makes studying interactive, personalized, and voice-driven. That’s where the idea was born — an AI-powered study companion that actually talks to you using your own notes.

What it does

Our web app creates an interactive, AI-powered teaching experience. Users begin by uploading their PDF notes, which are processed and understood by Google Gemini. Once the content is absorbed, Gemini takes on the role of a curious student. It initiates the session by asking questions based on the uploaded material, encouraging the user to explain and teach the concepts out loud. As the session continues, Gemini keeps asking new questions, pushing the user to recall and reinforce what they know until the end of the session. This approach transforms passive note reviewing into an active, engaging teaching exercise that enhances understanding and retention.

How we built it

We built it with React.js on the frontend with Fast API with python on the backend with Gemini Embedding and Text Generation. Users are upload a PDF on the frontend which is read by the Fast API endpoints in the backend which parses the PDF, uses Gemini's Embedding model to embed it into Vectors which is then stored in a vector database, Chromadb, for Gemini to use. Afterwards we go into the 'Teaching Session' where Gemini uses the provided PDF to ask a initial question which users are able to answer by voice which Gemini then uses as further context to ask more questions. This way the user is able to simulate a teaching environment for themselves to learn the topic.

Challenges we ran into

One of the biggest challenges was generating good questions for Gemini to ask the user. The context would often be reset or the embeddings would be wrongly read. This was due to when we started the app we used OpenAI embeddings but we soon found out Gemini Text generation was not super compatible with this. We were easily able to fix this by switching to the Gemini Embedding model.

Accomplishments that we're proud of

One of our biggest accomplishments was successfully creating a seamless flow between uploading notes, having Gemini understand the content, and turning it into a real-time Q&A teaching session. We’re proud of how we brought the concept of active recall to life in a completely new way — by flipping the traditional learning process and making the user the teacher.

Integrating voice interaction and maintaining context through a session was a technical challenge, but we’re proud of how intuitive and engaging the final experience feels. Most importantly, we turned an idea that started with “what if studying was more interactive?” into a functional tool in just a short hackathon window.

What we learned

We learned a lot about how to use vector databases along with embedding models and hope to use these technologies to make more context-based application that speeds the process as a middleman to these models.

What's next for Recallify

We want to expand the app by adding other modes of questions, having statistics of how much of the users teaching is correct along with additional tools to generate more interactive study materials such as flashcards, practice tests and much more. We want it to be an all-in-one interactive study tool

Built With

• fastapi
• gemini
• javascript
• python
• react