PSG AI: Personalized Study Guide

Inspiration

Being a student in Nigeria comes with a ton of “billing” as we call it, From the cost of books to the cost of materials from lecturers. Not everyone can afford to buy these materials, and not everyone is tech-savvy enough to source relevant information online. We saw the need to bridge the gap and make learning easy, and affordable for everyone.

PSG does not require technical training, anyone who uses a smartphone will comfortably be able to navigate their way through the app. And without the block of having to pay to learn, they can easily upload their curriculum and get the materials they need for free.

This ensures that everyone has access to quality educational materials which include generated AI content, online articles, and YouTube videos, at the click of a button, regardless of their financial standing or technical know-how.

What it does

Our project, Personalized Study Guide, is a web application designed to assist learners in creating organized course materials, topics, and resources derived from a syllabus to enhance the learning process.

How we built it

The app's user interface was built using Tailwind CSS, Sveltekit, and Shadcn-svelte.
The app uses Firebase for user management and authentication, Firestore for data storage, and FastApi for building the endpoints.
The app extracts topics from the syllabus uploaded as images using Gemini Pro vision and uses a custom-made OCR for PDF text extraction. The app leverages Gemini Pro API for content structuring, topic resource gathering, web scraping, and topic note generation.
The app has an intuitive, elegant, and user-friendly interface, making it accessible to users within diverse groups.
Created courses on the app are stored, and users don't have to regenerate these topics and resources until they want to.

Challenges we ran into

We had a challenge with API token limit exhaustion, but we implemented strategies to optimize our API calls and avoid reaching the limit.
Another challenge we had was Gemini Pro's output token limit, but we devised a solution to deliver complete information.

Accomplishments that we're proud of

Intuitive App Interface: Designed a user-friendly and intuitive app interface for learning, ensuring a seamless user experience for students.
Streamlined In-App Resource Gathering: Built an efficient in-app resource gathering system within the application, allowing users to find relevant text and video materials quickly.
Custom, Powerful OCR Integration: Developed a powerful, custom Optical Character Recognition (OCR) module for efficient PDF content extraction. This allows users to upload their syllabus in PDF format.

What we learned

We discovered that when creating an AI tool, focusing on the users' needs and experiences is key.
We also learned the benefits of user research, to empathize with users and create a useful product for them.
In addition to this, we learned the value of teamwork, brainstorming, effective collaboration, and proactiveness, to deliver stellar output and amazing results.

What's next for PERSONALIZED STUDY GUIDE AI

Enhancing Learning with Gamification: Integrate a gamified quiz feature within the app. This allows users to test their knowledge through practice questions and receive immediate feedback, fostering active learning and knowledge retention.

In-App Learning Community: Integrate a dedicated in-app learning community feature. This allows learners to:

Connect with peers: Foster a network of learners to share study tips, exchange resources, and motivate each other.
Contribute to a knowledge base: Encourage users to upload their learning syllabus and materials, creating a valuable shared resource library within the app.

We believe this product has the potential to address educational needs not just in Nigeria but globally. In the long term, we see immense value in collaborating with other teams and organizations to further develop this solution.

Built With

fastapi
gemini
python
sveltekit
typescript

Submitted to

Google AI Hackathon

Created by

I initiated the project idea and guided the overall design direction. It was my first time working on an education-focused AI solution, and I learned a lot about how LLMs can be applied to adaptive learning.

1. I proposed the concept of using AI to create adaptive study plans in education—a domain that is inherently collaborative and impactful.

2. I implemented the technical pipeline that integrates LLMs with user input to generate personalized study recommendations.

3. I presented our solution at a local hackathon in Lagos, where our team secured second place

David Olukolatimi
Machine Learning Engineer
I worked on the product ideation, user experience/interface design and prototyping of the Personalized Study Guide app. It was my first time designing an app that leverages Gemini API to provide AI solutions for Education Technology, yet, it was exciting and I learned a whole lot on the job.

Othniel Anselm
I worked on the backend using FastAPI and Langchain and I also contributed to some part of the frontend using Sveltekit and Firebase.

Sheyzi Einstein
I worked on the back end, it was my first time using Fastapi, Langchain, and Google Gemini API.

1. I worked on the pdf extraction pre-processing flow from the API endpoint using a combination of OCR and mymupdf to extract text from uploaded pdf depending on if the page is a screenshot or just text. I used Langchain and Google Gemini-pro to extract and group topics and sub-topics.

2. I worked on the summarize endpoint, which took in a note previously generated and a topic to provide a summary together with a list of important points

3. Worked on the quiz section on the backend, utilizing Gemini-pro to generate practice quizzes, hints and solutions

goodnews john
I worked on the front end, this was my first time working with Firebase, the documentation contained everything I needed, I had a nice developer experience.

1. I translated the figma designs into our web app using sveltekit, tailwindcss for and shadcn-svelte for styling, Typescript and JavaScript for logic and api consumption.

2. I integrated the api endpoints with the user interface.

3. I wrote custom utility functions for processing the data bit by bit to improve user experience.

4. I worked on storing and managing the topics and courses using the Firestore database, User authentication was done with firebase

Michael Festus