app.py

import streamlit as st
from PIL import Image
from pytesseract import pytesseract
import PyPDF2
import enum
import os
# Install Tesseract at runtime
if not os.path.exists("/usr/bin/tesseract"):
    os.system("apt-get update && apt-get install -y tesseract-ocr libtesseract-dev")
    
import re
from collections import defaultdict
import folium
from streamlit_folium import st_folium
from geopy.geocoders import Nominatim
from geopy.exc import GeocoderTimedOut
import wikipedia
from transformers import pipeline
from openai import OpenAI

# NVIDIA OpenAI API Setup
client = OpenAI(
    base_url="https://integrate.api.nvidia.com/v1",
    api_key="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"  # Replace with actual API key
)

# Load Named Entity Recognition (NER) Model
nlp = pipeline("ner", model="dbmdz/bert-large-cased-finetuned-conll03-english")

st.set_page_config(page_title="OCR & Historical Analysis", page_icon="📜", layout="wide")

# Custom Styling
def style_text(text):
    return f"""
    <div style='padding:10px;border-radius:10px;
                background-color:#e0e0e0;
                color:#333;
                font-weight:500;
                font-size:16px;'>
        {text}
    </div>
    """

def find_related_documents(query):
    try:
        search_results = wikipedia.search(query, results=5)
        links = [wikipedia.page(result).url for result in search_results]
        return links
    except Exception as e:
        return [f"Error retrieving related documents: {str(e)}"]

def geocode_location(location):
    geolocator = Nominatim(user_agent="streamlit_app")
    try:
        loc = geolocator.geocode(location, timeout=10)
        return (loc.latitude, loc.longitude) if loc else None
    except GeocoderTimedOut:
        return None

def generate_historical_context_nvidia(text):
    """Use NVIDIA OpenAI API to generate a structured, summarized historical context."""

    prompt_analysis = f"""
    Analyze the following text and provide a historical context. Identify:
    - Key historical events
    - Significant figures involved
    - The broader historical significance

    Text: {text}

    Provide a detailed response.
    """

    prompt_summary = """
    Summarize the historical context provided above in a concise and structured format:
    - Limit to 5 bullet points
    - Each bullet point should be under 100 words
    - Avoid unnecessary explanations or preamble—return only the summary
    """

    try:
        # Step 1: Generate Detailed Historical Context
        completion = client.chat.completions.create(
            model="deepseek-ai/deepseek-r1",
            messages=[
                {"role": "system", "content": "You are a historian providing detailed historical insights."},
                {"role": "user", "content": prompt_analysis}
            ],
            temperature=0.4,
            top_p=0.9,
            max_tokens=4096,
            stream=False
        )
        detailed_response = completion.choices[0].message.content.strip()

        # Step 2: Summarize the Historical Context **without Monologue**
        summary_completion = client.chat.completions.create(
            model="deepseek-ai/deepseek-r1",
            messages=[
                {"role": "system", "content": "You are an expert summarizer."},
                {"role": "user", "content": f"{detailed_response}\n\n{prompt_summary}"}
            ],
            temperature=0.4,
            top_p=0.9,
            max_tokens=2048,
            stream=False
        )
        
        # Extract only the structured summary
        summary_response = summary_completion.choices[0].message.content.strip()
        
        # Remove AI-generated explanations or redundant preamble
        clean_summary = re.sub(r"^.*?\n\n", "", summary_response, flags=re.DOTALL)

        return clean_summary if clean_summary else "No historical context found."

    except Exception as e:
        return f"Error retrieving AI-generated historical context: {str(e)}"

    
class OS(enum.Enum):
    Mac = 0
    Windows = 1

class Languages(enum.Enum):
    English = "eng"
    Filipino = "fil"
    Spanish = "spa"

class ImageReader:
    def __init__(self, os):
        if os == OS.Windows:
            pytesseract.tesseract_cmd = '/usr/bin/tesseract'

    def extract_text(self, image: Image, lang: Languages):
        extracted_text = pytesseract.image_to_string(image, lang=lang.value)
        return ' '.join(extracted_text.split())

    def extract_text_from_pdf(self, pdf_file, lang: Languages):
        pdf_reader = PyPDF2.PdfReader(pdf_file)
        text = "".join(page.extract_text() or "" for page in pdf_reader.pages)
        return text

    def extract_key_details(self, text):
        details = {"dates": set(), "names": set(), "locations": set()}
        date_pattern = r'\b(?:\d{1,2}[/\-]\d{1,2}[/\-]\d{2,4}|\d{4})\b'
        details['dates'] = set(re.findall(date_pattern, text))
        entities = nlp(text)
        
        for entity in entities:
            if "PER" in entity['entity']:
                details['names'].add(entity['word'])
            elif "LOC" in entity['entity']:
                details['locations'].add(entity['word'])
        
        return details

# UI Layout
st.title("📜 OCR & Historical Context Analyzer")
st.markdown("Extract text from images and PDFs, analyze named entities, and retrieve historical context.")

col1, col2 = st.columns([1, 2])

with col1:
    selected_os = st.selectbox("🖥️ Select your OS", [OS.Windows, OS.Mac], format_func=lambda x: x.name)
    selected_lang = st.selectbox("🌍 Select language", list(Languages), format_func=lambda x: x.name)
    uploaded_file = st.file_uploader("📂 Upload an image or PDF", type=["png", "jpg", "jpeg", "pdf"])

if uploaded_file:
    ir = ImageReader(selected_os)
    extracted_text = ""
    if uploaded_file.type in ["image/png", "image/jpeg"]:
        image = Image.open(uploaded_file)
        st.image(image, caption="Uploaded Image", use_column_width=True)
        extracted_text = ir.extract_text(image, selected_lang)
    else:
        extracted_text = ir.extract_text_from_pdf(uploaded_file, selected_lang)

    st.markdown("### 📝 Extracted Text:")
    st.markdown(style_text(extracted_text), unsafe_allow_html=True)

    key_details = ir.extract_key_details(extracted_text)
    st.markdown("### 🔍 Extracted Key Details")
    st.write(f"**📅 Dates:** {', '.join(key_details['dates']) if key_details['dates'] else 'None found'}")
    st.write(f"**👤 Names:** {', '.join(key_details['names']) if key_details['names'] else 'None found'}")
    st.write(f"**📍 Locations:** {', '.join(key_details['locations']) if key_details['locations'] else 'None found'}")
    
    combined_terms = ' '.join(key_details['dates'].union(key_details['locations']).union(key_details['names']))
    historical_context = generate_historical_context_nvidia(combined_terms)
    st.markdown("### 🏛️ Historical Context")
    st.markdown(style_text(historical_context), unsafe_allow_html=True)
    
    st.markdown("### 🌐 Search the Web")
    search_query = st.text_input("Enter a keyword or phrase:")
    if search_query:
        search_results = generate_historical_context_nvidia(search_query)
        st.markdown(style_text(search_results), unsafe_allow_html=True)
    
    related_docs = find_related_documents(combined_terms)
    st.markdown("### 📚 Related Historical Documents")
    for link in related_docs:
        st.markdown(f"[🔗 {link}]({link})")
    
    st.markdown("### 🗺️ Map of Key Locations")
    map_center = [10.0, 10.0]
    map_obj = folium.Map(location=map_center, zoom_start=2)
    
    for loc in key_details['locations']:
        coords = geocode_location(loc)
        if coords:
            folium.Marker(coords, popup=loc).add_to(map_obj)
    
    st_folium(map_obj, width=700, height=500)