Automated Summary and Questionnaire Generator

Description

Web application that processes files (PDF, DOCX, PPTX) or YouTube transcripts to generate:

• 📌 Automatic summaries using NLP (spaCy + TextRank)
• ❓ Interactive AI Quizzes (Call 3 70B via NVIDIA API)

Why this project?
This generator automates the creation of summaries and quizzes from study materials, helping students and teachers reduce study time and provide quick assessments based on existing content. It's ideal for situations where large amounts of text or audiovisual content need to be processed efficiently.

Who is it for?
The system is designed to be used by:

• Students looking to quickly review content using automatically generated summaries and quizzes.
• Teachers who want to generate personalized assessments or summaries of their recorded classes or teaching materials.

System Screenshots

1. Similarity Analysis (Histogram)

Optimal threshold: 0.85 (configurable in code)

The histogram shows how similarity analysis is performed to determine the relationship between text sections. A higher similarity threshold can improve the accuracy of summaries.

2. Main Interface

Selecting between local files or YouTube URLs

The interface is easy to use and allows users to upload local files or paste YouTube links directly to start generating summaries and quizzes.

3. Summary Example

Reduction from 728 words → 120 words (83% more concise)

The summary is generated using NLP techniques such as TextRank, allowing a considerable reduction in text length without losing the essence of the content.

4. Generated Questionnaire

The 5 questions with multiple choices and explanations can be modified in this section of code, where it must be considered that the greater the number of questions, the greater the cost in tokens of the query and therefore the number of requests to the API is reduced, so for this case 5 were used.

The generated quiz is interactive and allows users to assess their understanding of the material. It also offers detailed explanations for each answer.

Technologies Used

Description of Technologies:

• spaCy : Used for natural language processing (NLP). Its ability to work with large volumes of text allows for efficient summary generation.
• TextRank : Graph-based algorithm for extracting the most relevant phrases from text, used for summarizing.
• Llama 3 70B : NVIDIA's next-generation language model used to generate interactive, meaningful questions, powered by the NVIDIA API.
• NVIDIA API : Platform used to access the Llama 3 model and generate custom questions.

flowchart TD
    A[Entrada] -->|Archivo PDF/DOCX/PPTX| B(Extracción de Texto)
    A -->|URL de YouTube| C(Transcripción API)
    B --> D[Texto Procesado]
    C --> D
    D --> E{Modo Seleccionado}
    E -->|Generar Resumen| F[spaCy + TextRank]
    E -->|Generar Cuestionario| G[Llama3 70B\nvía NVIDIA API]
    F --> H[Resumen Automático\nReducción 80% palabras]
    G --> I[Cuestionario JSON\n5 preguntas con opciones]
    H --> J[(Salida:\nMarkdown/Interfaz)]
    I --> J
    K[Streamlit] -->|Interfaz Web| L[Usuario Final]

    %% Estilos
    classDef tech fill:#4CAF50,color:white,stroke:#388E3C;
    classDef data fill:#2196F3,color:white,stroke:#1976D2;
    classDef output fill:#FF9800,color:white,stroke:#F57C00;
    classDef tool fill:#9C27B0,color:white,stroke:#7B1FA2;

    class B,C,F,G,K tech;
    class D,A data;
    class H,I,J output;
    class L tool;

classDiagram
    class Streamlit {
        +file_uploader()
        +text_input()
        +button()
    }
    class spaCy {
        +load(model_name)
        +add_pipe(algorithm)
    }
    class NVIDIA_API {
        +base_url: string
        +model: string
    }
    Streamlit --> spaCy : Usa para
    Streamlit --> NVIDIA_API : Consulta

How to Use

1. Upload a file (PDF/DOCX/PPTX) or paste YouTube URL .

2. Choose the mode :

   • ✂️ Summary : Generates a summary based on text similarity analysis.
   • 📝 Quiz : Generates an interactive quiz based on the analyzed content.

3. Explore the results :

   • Summary : You can export it to Markdown or PDF to share or study.
   • Quiz : Allows you to answer interactive questions and check your understanding.

How are files processed?

• PDF/DOCX/PPTX files : The text is extracted using specific libraries such as PyPDF2 for PDFs and python-docx for Word documents. It is then processed to generate the summary or questionnaire.
• YouTube : The Transcription API extracts audio from the video and converts it to text. The text is then used to generate the summary or quiz.

Resources

Models

Datasets

• YouTube transcripts (via public API).
• Files uploaded by users.

Requirements:

• NVIDIA API key for Llama 3 : Required to access the model's capabilities.
• Internet connection : Required to get YouTube transcripts and use the NVIDIA API.

Grades

You need to download the Markdown Preview Mermaid Support extension.
- this is so that the diagrams are displayed correctly in mermaid format

⚠️ Limitations :

• Summary Accuracy : Although the system is designed to create accurate summaries, quality depends on the content and format of the text. Some details may be lost in the reduction.
• YouTube Restrictions : The system may have difficulty transcribing videos with access restrictions, such as private or copyrighted videos.

🛠️ Code available at : GitHub/repo