Project: GptScholar (Team Schulich)

[Pelumioluwa]

Project name

GptScholar

Description

GptScholar is an advanced multilingual AI chat application designed to support first-year undergraduate students with their academics. Using OpenAI's GPT-4 Turbo model, Azure AI Search and Streamlit for the interface, it provides tailored assistance and insights. The platform allows students to seamlessly upload PDF files, input URLs from webpages, or select specific subjects for guidance.

Data sources include various types including textbook PDFs, URL links, and other PDF documents. The application uses PDF documents containing valuable insights across subjects like Calculus, Physics, Computer Science, and Finance.

To enhance data ingestion, an ingest file was developed to extract and process relevant information from textbooks, push it to Cosmos DB, and retrieve information using Azure AI Search. This enabled efficient data retrieval and interaction with the Large Language Models (LLMs) in the RAG architecture. Further enhancements involve preprocessing steps to clean and standardize text content from uploaded PDFs and url links, ensuring optimal formatting for analysis by the GPT-4 Turbo model.

The application supports multilingual interactions, through the following media

• Typing queries: Users are able to input queries and receive responses in any of the following languages; English, Spanish, French, Hindi, Chinese, Arabic, Russian, Portuguese, Japanese, German, Korean, Italian, Turkish, and Dutch

• Speaking: Users are able to say their questions through its speech-to-text functionalities in any of the following languages; English, Spanish, Italian, and Hindi.

GptScholar caters to diverse users, including:

• Students: Providing tailored academic support, clarification, and exploration of course concepts.
• Researchers: Facilitating research, insights gathering, and topic clarification across industries.
• Language Learners: Assisting in language practice, comprehension, and writing skills development.
• Fact-Checkers: Offering quick access to reliable information for verification purposes.
• Individuals with disabilities: Through our speech-to-text functionality, individuals with disabilities are able to utilize the chatbot
• General Users: Enabling exploration and retrieval of information from pdf and url links conveniently.

Language

User query and response for GptScholar: English, Spanish, French, Hindi, Chinese, Arabic, Russian, Portuguese, Japanese, German, Korean, Italian, Turkish, and Dutch

Speech-to-text Languages: English, Spanish, Italian, and Hindi.

Project Repository URL

https://github.com/offcial17/RAG_Hacktogether

Deployed Endpoint URL

http://hackapp.eastus.azurecontainer.io/

Project video

https://www.loom.com/share/2624be81ca784023a486aacaad3a873f?sid=74fe5df5-114c-45f6-b19a-6ec7273b4fdc

Team members

Sabrina Renna: https://github.com/srenna, Sushmit Richard: https://github.com/offcial17, Pelumioluwa Abiola: https://github.com/Pelumioluwa

Showcase Consent

Yes

[srenna]

Love this post @Pelumioluwa ! Thanks for sharing our work.

[Pelumioluwa]

Footnote: Due to browser related issues with access permissions, speech-to-text features is only available when app is run locally