ChartTable-RAG / app.py
KhwaishK's picture
Update app.py
3fa6719 verified
Raw History Blame Contribute Delete
4.14 kB
# app.py
import os
import gradio as gr
from dotenv import load_dotenv
# Load secret API keys globally
load_dotenv()
# Import the functional pipelines you built!
from extraction import run_extraction_pipeline
from summarize import run_summarization_pipeline
from database import run_database_pipeline
from pipeline import create_rag_chain
# This global variable will hold our live pipeline once a file is processed
global_rag_chain = None
def process_uploaded_pdf(file_obj):
"""
Takes an uploaded file object from the UI, extracts the path,
runs it through your ingestion pipeline, and updates the active chain.
"""
global global_rag_chain
if file_obj is None:
return "❌ Please upload a valid PDF file first."
try:
# Gradio saves the uploaded file in a temporary folder.
# .name gives us the absolute string path to that file (e.g., '/tmp/xyz.pdf')
uploaded_file_path = file_obj.name
gr.Info("⏳ Step 1/3: Partitioning and chunking the document...")
tables, text_chunks, images = run_extraction_pipeline(uploaded_file_path)
gr.Info("⏳ Step 2/3: Summarizing layouts, text chunks, and images...")
text_sums, table_sums, image_sums = run_summarization_pipeline(tables, text_chunks, images)
gr.Info("⏳ Step 3/3: Indexing summaries into Multi-Vector store...")
live_retriever = run_database_pipeline(text_chunks, text_sums, tables, table_sums, images, image_sums)
# Build the final execution chain and update the global manager variable
global_rag_chain = create_rag_chain(live_retriever)
return "βœ… Document fully processed and indexed! You can now start asking questions on the right."
except Exception as e:
return f"❌ Pipeline Ingestion Failed: {str(e)}"
def predict(message, history):
"""
Handles user chat questions. Checks if a database is currently loaded before
running inference.
"""
global global_rag_chain
# Safeguard block if the user tries to chat before uploading a file
if global_rag_chain is None:
return "⚠️ No document database found. Please upload a PDF and click 'Process Document' on the left panel first."
if not message.strip():
return "Please enter a valid question."
try:
# Run the production source chain using the user's question string
result = global_rag_chain.invoke(message)
return result["response"]
except Exception as e:
return f"❌ An error occurred during inference extraction: {str(e)}"
# --- GRADIO INTERFACE ARCHITECTURE ---
# We use Blocks here to design a professional two-column application
with gr.Blocks() as demo:
gr.Markdown("# πŸ€– ChartTable RAG")
gr.Markdown("Upload **any** document containing text, data tables, and diagrams. The pipeline will build an isolated vector database on the fly.")
with gr.Row():
# --- LEFT PANEL: File Upload and Processing Management ---
with gr.Column(scale=1):
gr.Markdown("### πŸ“₯ Step 1: Ingestion Dashboard")
file_input = gr.File(label="Upload your PDF file", file_types=[".pdf"])
process_btn = gr.Button("βš™οΈ Process Document", variant="primary")
status_output = gr.Textbox(label="System Status Log", interactive=False, value="Awaiting file upload...")
# --- RIGHT PANEL: Interactive Multi-Modal Chat Engine ---
with gr.Column(scale=2):
gr.Markdown("### πŸ’¬ Step 2: Ask Your Data")
gr.ChatInterface(
fn=predict,
)
# Connect the left side click event button action to our processing function
process_btn.click(
fn=process_uploaded_pdf,
inputs=[file_input],
outputs=[status_output]
)
# Launch the app environment container
if __name__ == "__main__":
demo.launch(theme = "soft")