Project: AI Book Assistant - Help low resources students access and study from teachers' books.

[DeaMariaLeon]

Project Name

AI Book Assistant

Description

Application Description

We built a web-based application that provides an easy-to-use RAG to allow students to learn class material more quickly.

One of the benefits of using the RAG instead of having students use an LLM chatboat like chatGPT is that the answers are based solely on the course material and the risk of having the LLM hallucinate is thus minimal.

It can be accessed at http://ai-demo.fr

Please note that https access has not yet been activated and you should access the application with http only.

This submission demonstrates that it is possible to enhance class material at a very low-cost to the great benefit of students around the world.

From the point of view of the student, they have a very simple yet clear user interface to retrieve the content they need to study. They can ask a question in the web application and they will get :

• references to the relevant passages of the material they have to study

• an answer to their question based on the retrieved passages.

The emphasis is here on the cost effectiveness and ease-of-use of the system.

For demonstration purposes, the application is currently using two freely available javascript books. Questions should thus be about Javascript. If a question is not about Javascript or if the answer cannot be found in the source material, the RAG will answer that it does not know.

Answers end with a word of encouragement for the student to keep learning.

Challenge Topic

Education: Help low resources students access and study from teachers' books.

Cost analysis

This is meant as a low-cost solution for schools or institutions scaling to hundreds of students.

As the application is built on free Open Source components, there is no upfront cost to deploying the system.

The cost for the virtual private server is around 5$ per month, which is about 5 cents per student per month on the assumption of the systeme being used in an institution of 100 students.

For the LLM part of the RAG chain, assuming the use of the GPT 3.5 API at prices around 0.00015$ per 1K tokens, a use of around 500 tokens per request and 50 requests per student per day, the cost per student per month would thus be around 10 cents.

A ballpark estimate of the system cost (server + API usage) would thus be around 15 cents per student per month. The total cost for a school with 100 students would thus be around 15$ per month.

future work

• In the web application, add a sign up and logging in for students and teachers

• Add in the web application, the possibility for teachers to upload new materials

• Add the selection of different databases for different classes

• Try if a small LLM model could run in the server instead of using an LLM API

• Benchmarking load and response time of the server with increasing numbers of users

Build

Yes

Train

No

Analyze

No

Challenge Topic / Topic Category

• Customized Information Extraction
• Education
• Enhancing Accessibility in Healthcare Through AI
• Mitigating Natural Disasters in a Changing Climate
• Reducing Food Waste Through AI-Powered Innovation
• other - World Issues
• other - Open-Source
• other - Scientific Computing
• other

Project Repository URL

https://github.com/davidleon123/Book-Assistant.git

Deployed Endpoint URL

http://ai-demo.fr

Project Video File (not folder) Link (ensure viewer access)

https://drive.google.com/file/d/1aOG-6w_q0o40pn7BRT7aXSj9X1mwIMrh/view?usp=sharing

Team Members

@ davidleon123, @DeaMariaLeon

[naumnaum]

Your project is a great initiative that addresses a real need—helping students learn more effectively by focusing on course-specific material while minimizing the risk of LLM hallucinations.

Deploying a demo endpoint for public use is a significant achievement. It allows others to interact with your solution and see its value firsthand!

If cost reduction is on your roadmap, there are excellent alternatives to using small models. Hosting larger models like llama3-70B/400B on API providers or GPU hosting platforms such as Modal can be significantly cheaper than many mainstream LLM APIs. Plus, it eliminates infrastructure maintenance concerns, allowing you to concentrate on refining the product.

One area for improvement would be creating a more straightforward and well-documented repository structure. This would invite contributions from the open-source community and help your project grow faster with external input.

I also noticed the planned feature for ingesting new books, which is an excellent next step! Make sure to support a variety of formats (e.g., PDFs with or without graphics, DOCX, etc.) to maximize the versatility and accessibility of your system.

This is a practical, impactful project with strong potential to scale and benefit students globally. Well done!

[DeaMariaLeon]

Thank you so much for your feedback @naumnaum !