# -*- coding: utf-8 -*- """App Automatically generated by Colab. """ # Note: The eSpeak installation code has been removed. # Instead, ensure that "espeak" is listed in your apt.txt file for Hugging Face Spaces. import gradio as gr import numpy as np from transformers import pipeline import os import groq # Updated imports to address LangChain deprecation warnings: from langchain_groq import ChatGroq from langchain.schema import HumanMessage from langchain.text_splitter import RecursiveCharacterTextSplitter from langchain_community.vectorstores import Chroma from langchain_community.embeddings import HuggingFaceEmbeddings from langchain.docstore.document import Document # Importing chardet (make sure to add chardet to your requirements.txt) import chardet import fitz # PyMuPDF for PDFs import docx # python-docx for Word files import gtts # Google Text-to-Speech library from pptx import Presentation # python-pptx for PowerPoint files import re # Initialize Whisper model for speech-to-text transcriber = pipeline("automatic-speech-recognition", model="openai/whisper-base.en") # Set API Key (Ensure it's stored securely in an environment variable) groq.api_key = os.getenv("GROQ_API_KEY") # Replace with a valid API key # Initialize Chat Model chat_model = ChatGroq(model_name="DeepSeek-R1-Distill-Llama-70b", api_key=groq.api_key) # Initialize Embeddings embedding_model = HuggingFaceEmbeddings() # Initialize ChromaDB vectorstore = Chroma(embedding_function=embedding_model) # Prompt for quiz generation with added remark quiz_prompt = """ You are an AI assistant specialized in education and assessment creation. Given an uploaded document or text, generate a quiz with a mix of multiple-choice questions (MCQs) and fill-in-the-blank questions. The quiz should be directly based on the key concepts, facts, and details from the provided material. Remove all unnecessary formatting generated by the LLM, including tags, asterisks, markdown formatting, and any bold or italic text, as well as **, ###, ##, and # tags. For each question: - Provide 4 answer choices (for MCQs), with only one correct answer. - Ensure fill-in-the-blank questions focus on key terms, phrases, or concepts from the document. - Include an answer key for all questions. - Ensure questions vary in difficulty and encourage comprehension rather than memorization. - Additionally, implement an instant feedback mechanism: - When a user selects an answer, indicate whether it is correct or incorrect. - If incorrect, provide a brief explanation from the document to guide learning. - Ensure responses are concise and educational to enhance understanding. Output Example: 1. Fill in the blank: The LLM Agent framework has a central decision-making unit called the _______________________. Answer: Agent Core Feedback: The Agent Core is the central component of the LLM Agent framework, responsible for managing goals, tool instructions, planning modules, memory integration, and agent persona. 2. What is the main limitation of LLM-based applications? a) Limited token capacity b) Lack of domain expertise c) Prone to hallucination d) All of the above Answer: d) All of the above Feedback: LLM-based applications have several limitations, including limited token capacity, lack of domain expertise, and being prone to hallucination, among others. """ # Function to clean AI response by removing unwanted formatting def clean_response(response): """Removes tags, asterisks, and markdown formatting.""" cleaned_text = re.sub(r".*?", "", response, flags=re.DOTALL) cleaned_text = re.sub(r"(\*\*|\*)", "", cleaned_text) cleaned_text = re.sub(r"^#+\s*", "", cleaned_text, flags=re.MULTILINE) return cleaned_text.strip() # Function to generate quiz based on content def generate_quiz(content): prompt = f"{quiz_prompt}\n\nDocument content:\n{content}" response = chat_model([HumanMessage(content=prompt)]) cleaned_response = clean_response(response.content) return cleaned_response # Function to retrieve relevant documents from vectorstore based on user query def retrieve_documents(query): results = vectorstore.similarity_search(query, k=3) return [doc.page_content for doc in results] # Function to handle chatbot interactions def chat_with_groq(user_input): try: relevant_docs = retrieve_documents(user_input) context = "\n".join(relevant_docs) response = chat_model([HumanMessage(content=user_input + "\n\nContext:\n" + context)]) cleaned_response_text = clean_response(response.content) audio_file = speech_playback(cleaned_response_text) return cleaned_response_text, audio_file except Exception as e: return f"Error: {str(e)}", None # Function to play response as speech using gTTS def speech_playback(text): tts = gtts.gTTS(text, lang='en') audio_file = "output_audio.mp3" tts.save(audio_file) return audio_file # Function to detect encoding safely def detect_encoding(file_path): try: with open(file_path, "rb") as f: raw_data = f.read(4096) detected = chardet.detect(raw_data) encoding = detected["encoding"] return encoding if encoding else "utf-8" except Exception: return "utf-8" # Function to extract text from PDF def extract_text_from_pdf(pdf_path): try: doc = fitz.open(pdf_path) text = "\n".join([page.get_text("text") for page in doc]) return text if text.strip() else "No extractable text found." except Exception as e: return f"Error extracting text from PDF: {str(e)}" # Function to extract text from Word files (.docx) def extract_text_from_docx(docx_path): try: doc = docx.Document(docx_path) text = "\n".join([para.text for para in doc.paragraphs]) return text if text.strip() else "No extractable text found." except Exception as e: return f"Error extracting text from Word document: {str(e)}" # Function to extract text from PowerPoint files (.pptx) def extract_text_from_pptx(pptx_path): try: presentation = Presentation(pptx_path) text = "" for slide in presentation.slides: for shape in slide.shapes: if hasattr(shape, "text"): text += shape.text + "\n" return text if text.strip() else "No extractable text found." except Exception as e: return f"Error extracting text from PowerPoint: {str(e)}" # Function to process documents safely def process_document(file): try: file_extension = os.path.splitext(file.name)[-1].lower() if file_extension in [".png", ".jpg", ".jpeg"]: return "Error: Images cannot be processed for text extraction." if file_extension == ".pdf": content = extract_text_from_pdf(file.name) elif file_extension == ".docx": content = extract_text_from_docx(file.name) elif file_extension == ".pptx": content = extract_text_from_pptx(file.name) else: encoding = detect_encoding(file.name) with open(file.name, "r", encoding=encoding, errors="replace") as f: content = f.read() text_splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=50) documents = [Document(page_content=chunk) for chunk in text_splitter.split_text(content)] vectorstore.add_documents(documents) quiz = generate_quiz(content) return f"Document processed successfully (File Type: {file_extension}). Quiz generated:\n{quiz}" except Exception as e: return f"Error processing document: {str(e)}" # Function to handle speech-to-text conversion def transcribe_audio(audio): sr, y = audio if y.ndim > 1: y = y.mean(axis=1) y = y.astype(np.float32) y /= np.max(np.abs(y)) return transcriber({"sampling_rate": sr, "raw": y})["text"] # Gradio UI with gr.Blocks() as demo: gr.HTML("

AI Tutor

") gr.HTML("""
""") with gr.Row(): with gr.Column(): audio_input = gr.Audio(type="numpy", label="Record Audio") transcription_output = gr.Textbox(label="Transcription") user_input = gr.Textbox(label="Ask a question") chat_output = gr.Textbox(label="Response") audio_output = gr.Audio(label="Audio Playback") submit_btn = gr.Button("Ask") with gr.Column(): file_upload = gr.File(label="Upload a document") process_status = gr.Textbox(label="Processing Status", interactive=False) process_btn = gr.Button("Process Document") audio_input.change(fn=transcribe_audio, inputs=audio_input, outputs=transcription_output) transcription_output.change(fn=lambda x: x, inputs=transcription_output, outputs=user_input) submit_btn.click(chat_with_groq, inputs=user_input, outputs=[chat_output, audio_output]) process_btn.click(process_document, inputs=file_upload, outputs=process_status) # Launch the Gradio app demo.launch()