Hindi-Rag / app.py
wellwisherofindia's picture
Update app.py
014e300 verified
raw
history blame
19 kB
import gradio as gr
import os
import tempfile
import time
import uuid
from datetime import datetime
import fitz
import requests
import json
import numpy as np
from sentence_transformers import SentenceTransformer
import faiss
from groq import Groq
from gtts import gTTS
import subprocess
import warnings
warnings.filterwarnings("ignore")
CONFIG = {
'PASSCODE': os.getenv('PASSCODE'),
'MAX_FILE_SIZE': 10 * 1024 * 1024,
'MAX_QUERIES_PER_SESSION': 10,
'MAX_AUDIO_DURATION': 120,
'GROQ_API_KEY': os.getenv('GAPI'),
'AUDIO_CLIP_DURATION': 10,
'BOOK_THUMBNAILS_DIR': './book_thumbnails',
'OCR_BOOKS_DIR': './ocr_books',
}
SESSION_DATA = {
'authenticated': False,
'session_id': str(uuid.uuid4()),
'query_count': 0,
'document_chunks': [],
'faiss_index': None,
'author_name': '',
'book_title': '',
'embedding_model': None,
'groq_client': None
}
# Predefined questions for books
PREDEFINED_QUESTIONS = {
'general': [
"इस पुस्तक का मुख्य विषय क्या है?",
"लेखक ने इस पुस्तक में क्या संदेश दिया है?",
"इस पुस्तक में कौन से मुख्य पात्र हैं?"
],
'analysis': [
"इस पुस्तक की मुख्य शिक्षा क्या है?",
"लेखक की लेखन शैली कैसी है?",
"इस पुस्तक में कौन सा मुख्य संघर्ष है?"
],
'content': [
"इस कहानी का क्या अंत है?",
"पुस्तक में कौन सी मुख्य घटनाएं हैं?",
"मुख्य पात्र का चरित्र कैसा है?"
]
}
def load_models():
if SESSION_DATA['embedding_model'] is None:
print("Loading embedding model...")
SESSION_DATA['embedding_model'] = SentenceTransformer('paraphrase-multilingual-MiniLM-L12-v2')
if SESSION_DATA['groq_client'] is None:
if CONFIG['GROQ_API_KEY']:
print("Initializing Groq client...")
SESSION_DATA['groq_client'] = Groq(api_key=CONFIG['GROQ_API_KEY'])
else:
print("Warning: GROQ_API_KEY not found")
return SESSION_DATA['embedding_model'], SESSION_DATA['groq_client']
def trim_audio_to_duration(input_path, output_path, duration=10):
try:
cmd = [
'ffmpeg', '-i', input_path,
'-t', str(duration),
'-acodec', 'copy',
'-y',
output_path
]
result = subprocess.run(cmd, capture_output=True, text=True)
if result.returncode == 0:
return True
else:
print(f"FFmpeg error: {result.stderr}")
return False
except Exception as e:
print(f"Error trimming audio: {str(e)}")
return False
def transcribe_audio(audio_file):
if audio_file is None:
return ""
if not CONFIG['GROQ_API_KEY'] or SESSION_DATA['groq_client'] is None:
return "Error: Groq API key not configured"
try:
with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp_file:
trimmed_audio_path = tmp_file.name
if not trim_audio_to_duration(audio_file, trimmed_audio_path, CONFIG['AUDIO_CLIP_DURATION']):
print("Warning: Could not trim audio, using full duration")
trimmed_audio_path = audio_file
with open(trimmed_audio_path, "rb") as file:
transcription = SESSION_DATA['groq_client'].audio.transcriptions.create(
file=(os.path.basename(trimmed_audio_path), file.read()),
model="whisper-large-v3",
response_format="verbose_json",
language="hi"
)
if trimmed_audio_path != audio_file:
try:
os.unlink(trimmed_audio_path)
except:
pass
return transcription.text
except Exception as e:
try:
if 'trimmed_audio_path' in locals() and trimmed_audio_path != audio_file:
os.unlink(trimmed_audio_path)
except:
pass
return f"Transcription error: {str(e)}"
def text_to_speech(text):
if not text or len(text.strip()) == 0:
return None
try:
tts = gTTS(text=text, lang='hi', slow=False)
with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as tmp_file:
tts.save(tmp_file.name)
return tmp_file.name
except Exception as e:
print(f"TTS Error: {str(e)}")
return None
def extract_text_from_pdf(pdf_path):
text_content = ""
try:
pdf_document = fitz.open(pdf_path)
total_pages = len(pdf_document)
print(f"Processing PDF with {total_pages} pages...")
for page_num in range(total_pages):
page = pdf_document.load_page(page_num)
page_text = page.get_text()
if page_text.strip():
text_content += page_text + "\n"
pdf_document.close()
if not text_content.strip():
return "Error: No selectable text found in PDF. Please ensure the PDF contains selectable text, not just images."
return text_content
except Exception as e:
return f"Error extracting text: {str(e)}"
def extract_metadata(text):
lines = [line.strip() for line in text.split('\n')[:25] if line.strip()]
author_name = "अज्ञात लेखक"
book_title = "अनाम पुस्तक"
for i, line in enumerate(lines):
if any(word in line.lower() for word in ['लेखक', 'author', 'by', 'द्वारा', 'रचयिता']):
author_name = line
elif 10 < len(line) < 100 and not any(char.isdigit() for char in line[:20]):
if book_title == "अनाम पुस्तक":
book_title = line
return author_name, book_title
def chunk_text(text, chunk_size=400, overlap=50):
words = text.split()
chunks = []
for i in range(0, len(words), chunk_size - overlap):
chunk = ' '.join(words[i:i + chunk_size])
if chunk.strip():
chunks.append(chunk)
return chunks
def create_embeddings(chunks):
embedding_model, _ = load_models()
embeddings = embedding_model.encode(chunks, show_progress_bar=False)
dimension = embeddings.shape[1]
index = faiss.IndexFlatIP(dimension)
faiss.normalize_L2(embeddings)
index.add(embeddings.astype('float32'))
return index
def search_similar_chunks(query, top_k=3):
if SESSION_DATA['faiss_index'] is None or not SESSION_DATA['document_chunks']:
return []
embedding_model, _ = load_models()
query_embedding = embedding_model.encode([query], show_progress_bar=False)
faiss.normalize_L2(query_embedding)
scores, indices = SESSION_DATA['faiss_index'].search(query_embedding.astype('float32'), top_k)
results = []
for i, idx in enumerate(indices[0]):
if idx >= 0 and idx < len(SESSION_DATA['document_chunks']):
results.append({
'text': SESSION_DATA['document_chunks'][idx],
'score': float(scores[0][i])
})
return results
def call_groq_api(prompt, model="llama-3.1-8b-instant"):
if not CONFIG['GROQ_API_KEY'] or CONFIG['GROQ_API_KEY'] == 'your_groq_api_key_here':
return "⚠️ Groq API key not configured. Please set GROQ_API_KEY environment variable."
url = "https://api.groq.com/openai/v1/chat/completions"
headers = {
"Authorization": f"Bearer {CONFIG['GROQ_API_KEY']}",
"Content-Type": "application/json"
}
data = {
"model": model,
"messages": [{"role": "user", "content": prompt}],
"temperature": 0.7,
"max_tokens": 600
}
try:
response = requests.post(url, headers=headers, json=data, timeout=30)
response.raise_for_status()
return response.json()['choices'][0]['message']['content']
except Exception as e:
return f"Error calling LLM: {str(e)}"
def generate_rag_response(query, context_chunks):
if not context_chunks:
return "मुझे इस प्रश्न का उत्तर देने के लिए पर्याप्त जानकारी नहीं मिली।"
context = "\n\n".join([chunk['text'] for chunk in context_chunks])
prompt = f"""आप एक हिंदी पुस्तक सहायक हैं। निम्नलिखित जानकारी के आधार पर प्रश्न का उत्तर दें:
पुस्तक: {SESSION_DATA['book_title']}
लेखक: {SESSION_DATA['author_name']}
संदर्भ:
{context}
प्रश्न: {query}
निर्देश:
- हिंदी में संक्षिप्त और सटीक उत्तर दें
- उत्तर की शुरुआत में पुस्तक और लेखक का संदर्भ शामिल करें
- केवल दिए गए संदर्भ के आधार पर ही उत्तर दें
"""
response = call_groq_api(prompt)
return response
def authenticate(passcode):
if passcode == CONFIG['PASSCODE']:
SESSION_DATA['authenticated'] = True
return gr.update(visible=False), gr.update(visible=True), "✅ Welcome!"
else:
return gr.update(visible=True), gr.update(visible=False), "❌ Invalid passcode"
def process_document(pdf_file):
if pdf_file is None:
return "Please upload a PDF file", "", "", gr.update(visible=True), gr.update(visible=False), gr.update(choices=[])
try:
file_size = os.path.getsize(pdf_file.name)
if file_size > CONFIG['MAX_FILE_SIZE']:
return f"File too large! Max size: {CONFIG['MAX_FILE_SIZE'] // (1024*1024)}MB", "", "", gr.update(visible=True), gr.update(visible=False), gr.update(choices=[])
text_content = extract_text_from_pdf(pdf_file.name)
if not text_content.strip() or "Error" in text_content:
return text_content, "", "", gr.update(visible=True), gr.update(visible=False), gr.update(choices=[])
author_name, book_title = extract_metadata(text_content)
SESSION_DATA['author_name'] = author_name
SESSION_DATA['book_title'] = book_title
chunks = chunk_text(text_content)
SESSION_DATA['document_chunks'] = chunks
SESSION_DATA['faiss_index'] = create_embeddings(chunks)
SESSION_DATA['query_count'] = 0
# Generate predefined questions
questions = []
for category in PREDEFINED_QUESTIONS.values():
questions.extend(category)
success_msg = f"✅ Document processed successfully!"
return success_msg, book_title, author_name, gr.update(visible=False), gr.update(visible=True), gr.update(choices=questions[:6])
except Exception as e:
return f"Error processing document: {str(e)}", "", "", gr.update(visible=True), gr.update(visible=False), gr.update(choices=[])
def show_questions():
"""Show the questions section"""
return gr.update(visible=False), gr.update(visible=True)
def process_query(audio_input, text_input, predefined_question):
if SESSION_DATA['query_count'] >= CONFIG['MAX_QUERIES_PER_SESSION']:
return "⚠️ Query limit reached", None
if not SESSION_DATA['document_chunks']:
return "Please upload a document first", None
query_text = ""
# Priority: Predefined > Audio > Text
if predefined_question and predefined_question != "Select a question...":
query_text = predefined_question
elif audio_input:
query_text = transcribe_audio(audio_input)
if "error" in query_text.lower():
query_text = ""
if not query_text.strip() and text_input.strip():
query_text = text_input.strip()
if not query_text.strip():
return "Please ask a question", None
try:
similar_chunks = search_similar_chunks(query_text)
response_text = generate_rag_response(query_text, similar_chunks)
audio_response = text_to_speech(response_text)
SESSION_DATA['query_count'] += 1
formatted_response = f"**प्रश्न:** {query_text}\n\n**उत्तर:** {response_text}"
return formatted_response, audio_response
except Exception as e:
return f"Error processing query: {str(e)}", None
def reset_session():
SESSION_DATA.update({
'query_count': 0,
'document_chunks': [],
'faiss_index': None,
'author_name': '',
'book_title': '',
'session_id': str(uuid.uuid4())
})
return "✅ New session started!", "", "", gr.update(visible=True), gr.update(visible=False), gr.update(visible=False), gr.update(choices=[])
def create_interface():
with gr.Blocks(
title="Hindi Book Assistant",
theme=gr.themes.Soft(),
css="""
.main-container { max-width: 1200px; margin: 0 auto; }
.section-header { font-size: 1.2em; font-weight: bold; margin: 1em 0; }
.upload-area { border: 2px dashed #ccc; padding: 2em; text-align: center; margin: 1em 0; }
"""
) as demo:
gr.HTML("""
<div style="text-align: center; padding: 2em;">
<h1>📚 Hindi Book Assistant</h1>
<p>AI-powered assistant for Hindi books with voice support</p>
</div>
""")
# Authentication Section
with gr.Group(visible=True) as auth_section:
gr.Markdown("### 🔐 Enter Passcode")
passcode_input = gr.Textbox(
label="Passcode",
type="password",
placeholder="Enter access code..."
)
auth_button = gr.Button("🔓 Access", variant="primary")
auth_status = gr.Textbox(label="Status", interactive=False)
# Main Interface
with gr.Group(visible=False) as main_section:
# Step 1: Upload Document
with gr.Group(visible=True) as upload_section:
gr.Markdown("### 📄 Upload Your Book")
pdf_upload = gr.File(
label="Choose PDF file",
file_types=[".pdf"],
type="filepath"
)
process_btn = gr.Button("📖 Process Book", variant="primary", size="lg")
doc_status = gr.Textbox(label="Status", interactive=False)
# Step 2: Book Info (shown after processing)
with gr.Group(visible=False) as book_info_section:
gr.Markdown("### 📚 Book Information")
with gr.Row():
book_title_display = gr.Textbox(label="Book Title", interactive=False)
author_display = gr.Textbox(label="Author", interactive=False)
continue_btn = gr.Button("➡️ Continue to Questions", variant="primary", size="lg")
# Step 3: Ask Questions (shown after continue)
with gr.Group(visible=False) as query_section:
gr.Markdown("### 💬 Ask Questions About Your Book")
with gr.Tab("🎯 Quick Questions"):
predefined_dropdown = gr.Dropdown(
label="Choose a question",
choices=[],
value=None,
interactive=True
)
ask_predefined_btn = gr.Button("🔍 Ask This Question", variant="primary")
with gr.Tab("🎤 Voice Question"):
audio_input = gr.Audio(
label="Record your question (Hindi/English)",
sources=["microphone"],
type="filepath"
)
ask_voice_btn = gr.Button("🔍 Ask Voice Question", variant="primary")
with gr.Tab("⌨️ Type Question"):
text_input = gr.Textbox(
label="Type your question",
placeholder="Example: इस पुस्तक का मुख्य विषय क्या है?",
lines=2
)
ask_text_btn = gr.Button("🔍 Ask Text Question", variant="primary")
# Response Section
gr.Markdown("### 📝 Answer")
response_text = gr.Textbox(
label="Response",
lines=6,
interactive=False
)
response_audio = gr.Audio(
label="🔊 Audio Response",
interactive=False
)
# Reset Button
gr.Markdown("---")
reset_btn = gr.Button("🔄 Start New Session", variant="secondary")
# Event Handlers
auth_button.click(
authenticate,
inputs=[passcode_input],
outputs=[auth_section, main_section, auth_status]
)
process_btn.click(
process_document,
inputs=[pdf_upload],
outputs=[doc_status, book_title_display, author_display, upload_section, book_info_section, predefined_dropdown]
)
continue_btn.click(
show_questions,
outputs=[book_info_section, query_section]
)
ask_predefined_btn.click(
process_query,
inputs=[gr.State(None), gr.State(""), predefined_dropdown],
outputs=[response_text, response_audio]
)
ask_voice_btn.click(
process_query,
inputs=[audio_input, gr.State(""), gr.State("")],
outputs=[response_text, response_audio]
)
ask_text_btn.click(
process_query,
inputs=[gr.State(None), text_input, gr.State("")],
outputs=[response_text, response_audio]
)
reset_btn.click(
reset_session,
outputs=[doc_status, book_title_display, author_display, upload_section, book_info_section, query_section, predefined_dropdown]
)
demo.load(load_models)
return demo
def main():
print("🚀 Starting Hindi Book Assistant...")
print("📋 Loading AI models...")
load_models()
demo = create_interface()
print("✅ Ready!")
print(f"🔑 Passcode: {CONFIG['PASSCODE']}")
demo.launch(
share=True,
show_error=True,
)
if __name__ == "__main__":
main()