Textract

Textract empowers you to unlock the hidden value within your PDFs. Extract key information, summarize content, and find answers instantly. 📖

PDF Chatbot Documentation

This document provides an overview of the PDF Chatbot application, its architecture, workflow, and API endpoints. The project allows users to upload a PDF file, extract its content, and query the extracted text using a chatbot powered by the Llama-3.3-70B-Versatile model.

---

Project Workflow

1. Backend

• Built using FastAPI.
• Handles file uploads, PDF text extraction, and querying the Llama model for answers based on the uploaded content.

2. Llama Integration

• Utilizes Groq AI's interface with the Llama-3.3-70B-Versatile model for answering user queries.

3. Frontend

• Developed using React.js.
• Provides an intuitive user interface for uploading files and interacting with the chatbot.

---

Backend Code Overview

1. Extracting Text from PDF

The /upload/ endpoint handles PDF uploads and extracts text using the PyMuPDF library.

@app.post("/upload/")
async def upload_pdf(file: UploadFile = File(...)):
    # Check if the uploaded file is a PDF
    if not file.filename.endswith(".pdf"):
        raise HTTPException(status_code=400, detail=f"File {file.filename} is not a valid PDF.")

    try:
        # Read the uploaded file as bytes
        pdf_bytes = await file.read()

        # Open and process the PDF using PyMuPDF
        with fitz.open(stream=pdf_bytes, filetype="pdf") as doc:
            extracted_text = ""
            # Loop through all the pages in the PDF and extract the text
            for page_num in range(doc.page_count):
                page = doc[page_num]
                extracted_text += page.get_text("text") + "\n"  # Extract text in plain format

        # Save the extracted text in memory, using the filename as the key
        uploaded_pdf_text[file.filename] = extracted_text

        # Return a success message with the filename
        return {"message": "File uploaded and processed successfully.", "filename": file.filename}

    except Exception as e:
        # If something goes wrong, send an error response
        raise HTTPException(status_code=500, detail=str(e))

2. Querying the Llama Model

The /ask/ endpoint handles user queries based on the content of the uploaded PDF. It uses the Llama-3.3-70B-Versatile model for generating answers.

class AskRequest(BaseModel):
    filename: str  # The name of the uploaded file we want to query
    question: str  # The question to ask based on the file's content


@app.post("/ask/")
async def ask_question(data: AskRequest):
    # Check if the requested file is in our temporary storage
    if data.filename not in uploaded_pdf_text:
        raise HTTPException(status_code=400, detail="File not found or not yet uploaded.")

    try:
        # Get the extracted text for the requested file
        extracted_text = uploaded_pdf_text[data.filename]

        # Use the Llama handler to process the question and get an answer
        answer = get_answer_from_llama(extracted_text, data.question)

        # Return the question and the generated answer
        return {
            "question": data.question,
            "answer": answer,
        }

    except Exception as e:
        # If something goes wrong, return an error response
        raise HTTPException(status_code=500, detail=str(e))

3. Llama Handler Function

The get_answer_from_llama function interacts with the Llama model to generate answers.

def get_answer_from_llama(file_content: str, question: str) -> str:
    """
    This function asks the Llama model a question based on some input text and returns its answer.
    
    Arguments:
    - file_content: The text or content we want the Llama model to base its answer on.
    - question: The actual question we’re asking.

    Returns:
    - The answer generated by the Llama model as a string.
    """
    try:
        # Send a message to the Llama model with the content and question. 
        chat_completion = client.chat.completions.create(
            messages=[
                {
                    "role": "user",
                    "content": f"{file_content} Based on above text {question}",
                }
            ],
            model="llama-3.3-70b-versatile",
        )
        return chat_completion.choices[0].message.content
    except Exception as e:
        raise RuntimeError(f"Error in Groq client: {e}")

Frontend Code Overview

1.File Upload

The uploadFile function handles file uploads to the backend.

const uploadFile = async (file) => {
    const formData = new FormData(); // Create a FormData object for file upload
    formData.append('file', file); // Append the file to the FormData object

    try {
        // Send the file to the backend using Axios
        const response = await axios.post('http://localhost:8000/upload/', formData, {
            headers: {
                'Content-Type': 'multipart/form-data', // Set the appropriate header for file uploads
            },
        });
        console.log('Response:', response.data); // Log the server response for debugging
    } catch (error) {
        // Handle any errors during the file upload process
        console.log('Error Uploading File: ', error.response?.data || error.message);
    }
};

2.Sending a Question

The sendQuestion function sends a question to the backend and retrieves the answer.

const sendQuestion = async (inputValue) => {
    // Check if a file is uploaded
    if (!uploadedFileName) {
        alert("Please upload the file first!"); // Alert the user if no file is uploaded
        return;
    }
    try {
        // Send the user's question and filename to the backend via POST request
        const response = await axios.post("http://localhost:8000/ask/", {
            filename: uploadedFileName,
            question: inputValue,
        });

        console.log("Response: ", response.data); // Log the response for debugging
        setAnswer(response.data.answer); // Set the AI's answer using the `setAnswer` function
    } catch (error) {
        // Log any error that occurs during the request
        console.log("Error in sending message: ", error.response?.data || error.message);
    }
};

API Endpoints

1. Upload PDF

• URL: /upload/
• Method: POST
• Payload: multipart/form-data containing the file.
• Response:

  {
      "message": "File uploaded and processed successfully.",
      "filename": "example.pdf"
  }

  2. Ask a Question

• URL: /ask/
• Method: POST
• Payload:
  • filename: The name of the uploaded PDF file to query.
  • question: The question to ask based on the file's content.

  {
      "filename": "example.pdf",
      "question": "What is the content of the first page?"
  }

Dependencies

Backend:

• FastAPI: Framework for building APIs.
• PyMuPDF: Library for working with PDF files.
• pydantic: Data validation and settings management.
• Groq client: Interface for interacting with the Llama AI model.

Frontend:

• React.js: Frontend library for building user interfaces.
• Axios: HTTP client for making API requests.