Blame for app.py

import os

import gradio as gr
import pandas as pd
from geolocation_apis import addresses_to_coordinates, coordinates_to_route
from pdf_parser import process_pdf_hebrew_rtl
from llm_api import llm_parser
from map_visualize import generate_route_map


def summarize_items(items):
    """
    Returns a preview list for the dataset component.
    """
    return [
        [
            f"{i + 1}. [{item['source']}] " + (
                os.path.basename(item['file'].name) if item['source'] == 'PDF' else item['content'][:40]
            ) + "..."
        ]
        for i, item in enumerate(items)
    ]


def add_files(files, items):
    """
    Adds uploaded PDF file objects to the state.
    """
    new_items = [{"source": "PDF", "file": f} for f in files]
    items += new_items
    return items, gr.update(samples=summarize_items(items))


def add_text(text, items):
    """
    Adds pasted text content to the state.
    """
    if text.strip():
        items.append({"source": "Paste", "content": text.strip()})
    return items, gr.update(samples=summarize_items(items))


def delete_item(evt: gr.SelectData, items):
    """
    Deletes an item when selected in the dataset component.
    """
    del items[evt.index]
    return items, gr.update(samples=summarize_items(items))


def run_pipeline(items):
    """
    Processes PDFs and pasted text through the LLM pipeline,
    geocodes addresses, computes an optimized route,
    and writes two HTML outputs (data table and map).
    """
    # Separate file-based and pasted inputs
    pdf_paths = [item['file'].name for item in items if item['source'] == 'PDF']
    # Parse PDFs to text
    text_docs = [process_pdf_hebrew_rtl(path) for path in pdf_paths]
    # Include any pasted text directly
    pasted_texts = [item['content'] for item in items if item['source'] == 'Paste']

    # Combine for LLM parsing
    docs = text_docs + pasted_texts
    results, llm_parsed = llm_parser(docs=docs)

    # Geolocation and routing
    locations = addresses_to_coordinates([r.business_address for r in results])
    points = [loc.coords for loc in locations]
    route = coordinates_to_route(points)

    # Build data table
    df = pd.DataFrame([loc.to_dict() for loc in locations])
    llm_parsed = llm_parsed.rename(columns={'business_address': 'Address'})
    df = df.join(llm_parsed.set_index('Address'), on='Address')

    # Write HTML outputs
    data_output_path = './data_output.html'
    with open(data_output_path, mode='w', encoding='utf-8') as html:
        html.write(df.to_html(index=False, escape=False))

    map_output_path = './map_output.html'
    generate_route_map(route, map_output_path)

    return data_output_path, map_output_path


# Build Gradio interface
with gr.Blocks() as demo:
    state = gr.State([])

    with gr.Row():
        file_input = gr.Files(file_types=['.pdf'], label="Upload PDFs")
        file_btn = gr.Button("Add PDFs")

    with gr.Row():
        paste_box = gr.Textbox(lines=3, placeholder="Paste text content...", label="Paste Text")
        paste_btn = gr.Button("Add Text")

    gr.Markdown("### 📝 Inputs:")
    item_list = gr.Dataset(
        components=[gr.Textbox(visible=False)],  # hide the textbox
        samples=[],
        headers=["Items"]
    )

    # Register callbacks for managing inputs
    file_btn.click(add_files, inputs=[file_input, state], outputs=[state, item_list])
    paste_btn.click(add_text, inputs=[paste_box, state], outputs=[state, item_list])
    item_list.select(delete_item, inputs=state, outputs=[state, item_list])

    # Processing and output
    process_btn = gr.Button("Process & Download HTML")
    data_out = gr.File(label="📊 Business Data HTML")
    map_out = gr.File(label="🗺️ Route Map HTML")
    process_btn.click(run_pipeline, inputs=state, outputs=[data_out, map_out])

# Launch the app
if __name__ == '__main__':
    demo.launch()