NCTCMumbai's picture
Update app.py
6654000 verified
# APP.PY
from msal import PublicClientApplication
import requests
import gradio as gr
import pandas as pd
import tiktoken
import tempfile
from PyPDF2 import PdfReader
from tqdm import tqdm
from pydantic import BaseModel, Field
from phi.agent import Agent, RunResponse
from phi.model.groq import Groq
from sentence_transformers import SentenceTransformer
from sentence_transformers import CrossEncoder
#from gradio_client import Client, handle_file
import os
from pptx import Presentation
from pptx2img import PPTXConverter # For splitting slides
import uuid
import shutil
from PIL import Image
import pandas as pd
import requests
import gradio as gr
from pydantic import BaseModel, Field
from typing import List
import tiktoken
from datetime import datetime
import zipfile
from PIL import Image
import gradio as gr
import threading
import time
# Importing functions from files
# from upload_function import process_presentation,get_folder_id
# from view_ppt import search_ppts
# from stats_dashboard import get_dashboard_stats ,update_dashboard
# from search_slides import search_slides,combine_slides_as_zip
# Configure Microsoft Authentication
# Access secrets securely
GROQ_API_KEY = os.getenv("GROQ_API_KEY")
CLIENT_ID = os.getenv("CLIENT_ID")
TENANT_ID = os.getenv("TENANT_ID")
ADMIN_USERNAME = os.getenv("ADMIN_USERNAME")
ADMIN_PASSWORD = os.getenv("ADMIN_PASSWORD")
AUTHORITY = f"https://login.microsoftonline.com/{TENANT_ID}"
SCOPES = ["Files.ReadWrite.All", "User.Read"]
os.environ["GROQ_API_KEY"] = GROQ_API_KEY
embedding_model = SentenceTransformer('nomic-ai/nomic-embed-text-v1', trust_remote_code=True)
from sentence_transformers import CrossEncoder
cross_encoder = CrossEncoder('cross-encoder/ms-marco-MiniLM-L-6-v2') # For reranking) # For reranking
access_token_state = {"token": None}
flow_state = {"flow": None}
global headers
global df
global search_results
from config import temp_file_path # Import the global variable
headers = {
"Authorization": None,
"Content-Type": "application/json"
}
# Local cache directory for downloaded files
LOCAL_CACHE_DIR = "local_cache"
os.makedirs(LOCAL_CACHE_DIR, exist_ok=True)
app = PublicClientApplication(client_id=CLIENT_ID, authority=AUTHORITY)
# Define Metadata Schema
class PPTMetadata(BaseModel):
PPT_Unique_ID: str = Field(description="A unique identifier for the presentation (e.g., filename or hash).")
Suitable_Title: str = Field(description="A concise and meaningful title for the presentation.")
Slide_Category: str = Field(description="The category or theme of the slides (e.g., Risk management, Data Analytics, Technology etc ).")
PPT_Owner:str = Field(description="The owner of the presentation ie who makes the presentation (eg: NCTC,DG Systems, Directorate of Logistics etc ,Not available if not found )")
Audience_Forum: str = Field(description="The intended audience or forum for the presentation/to whom the presentaiton is made (e.g., NACIN, WCO, Presentation before Member (CBIC),Not available if not found).")
Short_Summary: str = Field(description="A brief summary of the presentation's content with all keywords in 10 sentences covering all keywords.")
# Function to download metadata file from OneDrive
def download_metadata_file(metadata_folder_id, headers):
metadata_file_name = "Master_metadata.csv"
url = f"https://graph.microsoft.com/v1.0/me/drive/items/{metadata_folder_id}/children"
response = requests.get(url, headers=headers)
if response.status_code != 200:
raise ValueError(f"Failed to list folder contents. Error: {response.text}")
items = response.json().get("value", [])
file_item = next((item for item in items if item['name'] == metadata_file_name), None)
if not file_item:
raise FileNotFoundError(f"{metadata_file_name} not found in OneDrive folder.")
download_url = file_item["@microsoft.graph.downloadUrl"]
response = requests.get(download_url)
if response.status_code != 200:
raise ValueError(f"Failed to download {metadata_file_name}. Error: {response.text}")
# Use tempfile to create a temporary file
with tempfile.NamedTemporaryFile(delete=False, suffix=".csv") as temp_file:
temp_file.write(response.content)
temp_file_path = temp_file.name # Save the path to the temporary file
print(f"βœ… Downloaded: {metadata_file_name} to temporary file: {temp_file_path}")
# with open(metadata_file_name, 'wb') as f:
# f.write(response.content)
# print(f"βœ… Downloaded: {metadata_file_name}")
return temp_file_path
##################################################### STATS DASHBOARD ##################################################################
def update_dashboard():
total_ppts, total_slides, chart_data, latest_html = get_dashboard_stats()
return (
gr.update(visible=True),
gr.update(value=f"<div><h3>Total PPTs: {total_ppts}</h3></div>"),
gr.update(value=f"<div><h3>Total Slides: {total_slides}</h3></div>"),
gr.update(value=chart_data),
gr.update(value=latest_html)
)
import pandas as pd
import gradio as gr
import os
def get_dashboard_stats():
# Load metadata CSV
global temp_file_path
global df
print('Reading CSV...',temp_file_path)
#metadata_file_name= "Master_metadata.csv"
# df = pd.read_csv(metadata_file_name)
#temp_file_path = os.path.join("/tmp", metadata_file_name)
df = pd.read_csv(temp_file_path)
# Ensure upload_date column is in datetime format
df["Upload_date"] = pd.to_datetime(df["Upload_date"], errors="coerce")
print(df)
# Total unique PPTs and slides
total_ppts = df["PPT_Unique_ID"].nunique()
total_slides = len(df)
# Monthly PPT uploads
df["month_year"] = df["Upload_date"].dt.to_period("M").astype(str)
monthly_stats = df.groupby("month_year")["PPT_Unique_ID"].nunique().reset_index()
monthly_stats.columns = ["Month", "PPT Uploads"]
# Gradio BarPlot requires a DataFrame
chart_data = monthly_stats
# Latest 5 PPTs by upload date
latest_df = df.drop_duplicates(subset="PPT_Unique_ID").sort_values("Upload_date", ascending=False)
latest_5 = latest_df[["Suitable_Title", "Slide_Category","Upload_date"]].head(5)
# Create HTML for the latest PPTs list
# Create HTML for the latest PPTs list with heading
latest_html = "<h4 style='margin-bottom: 8px;'>πŸ“Œ Top 5 Latest Uploaded PPTs</h4><ul style='line-height:1.6em;'>"
for _, row in latest_5.iterrows():
title = row["Suitable_Title"]
category = row["Slide_Category"]
date_str = row["Upload_date"].strftime("%Y-%m-%d") if pd.notnull(row["Upload_date"]) else "Unknown Date"
latest_html += f"<li><b>{title}</b> <br><i>{category}</i> β€” <span style='color:gray;'>{date_str}</span></li>"
latest_html += "</ul>"
return total_ppts, total_slides, chart_data, latest_html
############################################################# UPLOAD PPT #######################################################################
import requests
def get_access_token():
flow = app.initiate_device_flow(scopes=SCOPES)
print("Go to", flow["verification_uri"])
print("Enter the code:", flow["user_code"])
result = app.acquire_token_by_device_flow(flow)
if "access_token" not in result:
print("❌ Could not acquire token:", result.get("error_description"))
exit()
return result["access_token"]
# Function to generate a unique PPT ID
def generate_unique_ppt_id():
return str(uuid.uuid4())[:8] # Generate an 8-character unique ID
def truncate_text_to_tokens(text, max_tokens, model_name="cl100k_base"):
encoding = tiktoken.get_encoding(model_name)
tokens = encoding.encode(text)
truncated_tokens = tokens[:max_tokens]
return encoding.decode(truncated_tokens)
def split_and_convert_ppt(file_path, output_folder_slides, output_folder_images):
os.makedirs(output_folder_slides, exist_ok=True)
os.makedirs(output_folder_images, exist_ok=True)
presentation = Presentation(file_path)
slide_texts = []
file_name = os.path.basename(file_path).split('.')[0]
print('File Name ',file_name)
print('File Path ',file_path)
for i in range(len(presentation.slides)):
unique_slide_id = f"{file_name}_{ppt_unique_id}_slide_{i + 1}"
slide_file_path = os.path.join(output_folder_slides, f"{unique_slide_id}.pptx")
print('Slide_file_path',slide_file_path)
image_path = os.path.join(output_folder_images, f"{unique_slide_id}_slide_1.png") # refer to pptx2img it stores iamge in this format new_name = f"{pptx_name}_slide_{idx + 1}.png"
print('Image file path',image_path)
# βœ… Step 1: Create a single-slide PPTX
new_presentation = Presentation(file_path)
slide_indexes_to_remove = [j for j in range(len(new_presentation.slides)) if j != i]
for idx in sorted(slide_indexes_to_remove, reverse=True):
r_id = new_presentation.slides._sldIdLst[idx].rId
new_presentation.part.drop_rel(r_id)
del new_presentation.slides._sldIdLst[idx]
new_presentation.save(slide_file_path)
del new_presentation
# βœ… Step 2: Convert the single-slide PPTX to image
converter = PPTXConverter()
converter.convert_pptx_to_images(slide_file_path, output_folder_images)
print(f"Slide {i+1} converted to image: {image_path}")
# βœ… Step 3: Extract text from the slide image # Switching off OCR
#slide_text = extract_text_from_image(image_path)
#using PPTX for text extraction(actualy its quality is better then tesseratct)
# Extract text using python-pptx (editable text)
slide = presentation.slides[i]
pptx_text = ""
for shape in slide.shapes:
if hasattr(shape, "text"):
pptx_text += shape.text.strip() + "\n"
print(f"πŸ”‘ PPTX Text Extractedfrom slide {i + 1}:\n", pptx_text.strip())
slide_texts.append(pptx_text.strip())
return slide_texts
def generate_metadata_with_retry(full_text, retries=3, max_tokens=5000, decrement=100, model_name="cl100k_base"):
for attempt in range(1, retries + 2):
try:
truncated_text = truncate_text_to_tokens(full_text, max_tokens, model_name)
print(f"πŸ” Attempt {attempt}: Generating metadata with ~{count_tokens(truncated_text)} tokens...")
metadata = generate_metadata(truncated_text)
print("πŸ“ Metadata generated successfully.")
return metadata # βœ… Return on success
except Exception as e:
print(f"❌ Error on attempt {attempt}: {str(e)}")
if attempt == retries + 1:
print("🚨 Max retries reached. Metadata generation failed.")
return None
else:
max_tokens -= decrement
print(f"πŸ”„ Retrying with {max_tokens} tokens...")
# Function to generate metadata using phidata agent
def generate_metadata(ocr_text):
# Initialize the Agent with detailed instructions
metadata_agent = Agent(
name="Metadata Generator",
role="Generates structured metadata for presentations based on their content.",
instructions=[
"Your task is to analyze the provided text and generate structured metadata for the presentation.",
"Carefully evaluate the content to determine the most appropriate values for each metadata field.",
# Rule 1: PPT Unique ID
"For the 'PPT_Unique_ID', use the first 8 characters of the MD5 hash of the input text. "
"This ensures uniqueness across presentations.",
# Rule 2: Suitable Title
"For the 'Suitable_Title', create a concise and meaningful title that captures the essence of the presentation. "
"Focus on first slide where title of presentation is given along with key themes, topics, or keywords mentioned in the text.",
# Rule 3: Slide Category
"For the 'Slide_Category', classify the presentation into one of the following categories: "
"The category or theme of the slides (e.g., Risk management , Data Analytics , Technology etc)"
"Base your decision on the overall theme or subject matter of the content.",
# Rule 4 :PPT owner
"Find The owner of the presentation ie who makes the presentation (eg: Done by name and designation ie Mr. baswaraj ,Princpial ADG , Additional Director ,or organisations like NCTC,DG Systems, Directorate of Logistics etc)"
"Dont Asssume if u could not found ,mention Not Available"
# Rule 5: Audience/Forum
"For the 'Audience_Forum', identify the target audience or forum for the presentation. "
"(e.g.,NACIN , WCO, Presentation before Member (CBIC)etc )."
"Dont Asssume if could not found ,mention Not Available"
"Consider the tone, language, and purpose of the content.",
# Rule 6: Short Summary
"For the 'Short_Summary', provide a brief summary of the presentation's content with all keywords in 10 sentences. "
"Highlight the keywords ,topics, main points or objectives of the presentation.",
"Mention the title also in the short summary ,owner and audience of the presentation"
# General Guidelines
"Ensure all fields are filled and meaningful. If unsure about a field, make an educated guess based on the context.",
],
model=Groq(id="deepseek-r1-distill-llama-70b"), # Replace with actual model ID
response_model=PPTMetadata,
markdown=True,
debug_mode=True,
show_tool_calls=True,
monitoring=True)
# Run the agent to generate metadata
response = metadata_agent.run(
f"Generate data fields for the following presentation content: {ocr_text}")
return response.content
# Function to get folder ID in OneDrive
def get_folder_id(folder_path, headers):
folders = folder_path.split("/")
parent_id = None
print("creating folder id for ",folder_path)
for folder_name in folders:
url = f"https://graph.microsoft.com/v1.0/me/drive/root/children" if not parent_id else f"https://graph.microsoft.com/v1.0/me/drive/items/{parent_id}/children"
response = requests.get(url, headers=headers)
if response.status_code != 200:
print(f"Failed to retrieve folder '{folder_name}'. Error: {response.text}")
return None
items = response.json().get("value", [])
folder_item = next((item for item in items if item["name"] == folder_name and "folder" in item), None)
if not folder_item:
# Create the folder if it doesn't exist
create_url = "https://graph.microsoft.com/v1.0/me/drive/root/children" if not parent_id else f"https://graph.microsoft.com/v1.0/me/drive/items/{parent_id}/children"
create_response = requests.post(create_url, headers=headers, json={
"name": folder_name,
"folder": {},
"@microsoft.graph.conflictBehavior": "rename"
})
if create_response.status_code not in [200, 201]:
print(f"Failed to create folder '{folder_name}'. Error: {create_response.text}")
return None
folder_item = create_response.json()
parent_id = folder_item["id"]
return parent_id
# Function to upload file to OneDrive
def upload_to_onedrive(file_path, folder_id, headers):
file_name = os.path.basename(file_path)
upload_url = f"https://graph.microsoft.com/v1.0/me/drive/items/{folder_id}:/{file_name}:/content"
with open(file_path, "rb") as file:
file_content = file.read()
response = requests.put(upload_url, headers=headers, data=file_content)
if response.status_code in [200, 201]:
print(f"Uploaded {file_name} to OneDrive.")
return response.json()["id"]
else:
print(f"Failed to upload {file_name}. Error: {response.text}")
return None
# Function to count tokens using tiktoken
def count_tokens(text, model_name="cl100k_base"):
encoding = tiktoken.get_encoding(model_name)
tokens = encoding.encode(text)
return len(tokens)
def list_folder_files(folder_id, headers):
url = f"https://graph.microsoft.com/v1.0/me/drive/items/{folder_id}/children"
response = requests.get(url, headers=headers)
if response.status_code != 200:
raise ValueError(f"Failed to list folder contents. Error: {response.text}")
return response.json().get("value", [])
def download_onedrive_file(file_id, filename, headers):
url = f"https://graph.microsoft.com/v1.0/me/drive/items/{file_id}"
r = requests.get(url, headers=headers).json()
download_url = r.get("@microsoft.graph.downloadUrl")
response = requests.get(download_url)
with open(filename, 'wb') as f:
f.write(response.content)
def update_and_upload_metadata_simplified(metadata_list, metadata_folder_id, metadata_with_fulltext_folder_id, headers):
df_new = pd.DataFrame(metadata_list, columns=[
"Unique_Slide_ID", "Slide_OCR_Text", "PPT_OCR_Text", "Slide_Embedding", "Short_Summary_Embedding",
"PPT_Unique_ID", "Suitable_Title", "Slide_Category", "PPT_Owner", "Audience_Forum", "Short_Summary",
"Slide_File_Path", "Slide_File_ID", "Full_PPT_File_Path", "Full_PPT_File_ID",
"Thumbnail_File_Path", "Thumbnail_File_ID","Upload_date"])
for csv_file, folder_id, drop_column in [
("Master_metadata.csv", metadata_folder_id, 'PPT_OCR_Text'),
("Master_fulltext_metadata.csv", metadata_with_fulltext_folder_id, None)]:
#folder_id = get_folder_id(folder_path, headers)
files = list_folder_files(folder_id, headers)
file_item = next((item for item in files if item['name'] == csv_file), None)
print('File items', file_item)
if file_item:
download_onedrive_file(file_item['id'], csv_file, headers)
df_existing = pd.read_csv(csv_file)
df_merged = pd.concat([df_existing, df_new], ignore_index=True)
else:
df_merged = df_new
if drop_column:
df_merged = df_merged.drop(columns=[drop_column])
df_merged.to_csv(csv_file, index=False)
upload_to_onedrive(csv_file, folder_id, headers)
print(f"βœ… Uploaded: {csv_file}")
return "βœ…PPT Processing and Metadata update complete!"
# Main processing function
def process_presentation(file):
try:
# Step 0: Validate file format
file_path = file.name if hasattr(file, "name") else file
file_extension = os.path.splitext(file_path)[-1].lower()
gr.Info()
if file_extension not in ['.pptx']:
raise ValueError("Unsupported file format. Please upload .pptx")
# Extract the base file name (without extension)
file_name = os.path.basename(file_path).split('.')[0]
print('File Name ',file_name)
# Step 1: Generate unique PPT ID
global ppt_unique_id
ppt_unique_id = generate_unique_ppt_id()
upload_date = datetime.now().strftime('%Y-%m-%d')
# Step 2: Acquire access token via device flow
# access_token = get_access_token()
# print('access_token',access_token)
print('PPT_unique id',ppt_unique_id)
# Step 3: Get folder IDs for OneDrive
# headers = {
# "Authorization": f"Bearer {access_token}",
# "Content-Type": "application/json"
# }
gr.Info('Connecting to OneDrive..')
ppt_repo_folder_id = get_folder_id("Projects Apps/PPT Maker/ppt_repo", headers)
slides_repo_folder_id = get_folder_id("Projects Apps/PPT Maker/slides_repo", headers)
slide_image_repo_folder_id = get_folder_id("Projects Apps/PPT Maker/slide_image_repo", headers)
metadata_folder_id=get_folder_id('Projects Apps/PPT Maker/Metadata_file',headers)
metadata_with_fulltext_folder_id=get_folder_id('Projects Apps/PPT Maker/Metadata_with_fulltext',headers)
print('ppt_repo_folder_id',ppt_repo_folder_id)
print('slides_repo_folder_id',slides_repo_folder_id)
print('slide_image_repo_folder_id',slide_image_repo_folder_id)
print('metadata_folder_id',metadata_folder_id)
if not (ppt_repo_folder_id and slides_repo_folder_id and slide_image_repo_folder_id and metadata_folder_id) :
gr.Error('Could not find or create required folders in OneDrive.')
raise ValueError("Could not find or create required folders in OneDrive.")
# Step 2: Upload the full PPT file to OneDrive
#ppt_repo_folder_id = get_folder_id("Projects Apps/PPT Maker/ppt_repo")
# βœ… Step: Check if file already exists in ppt_repo
existing_files = list_folder_files(ppt_repo_folder_id, headers)
ppt_file_name = os.path.basename(file_path)
if any(item['name'] == ppt_file_name for item in existing_files):
gr.Error('⚠️ A file named ' + ppt_file_name + ' already exists in the PPT repository. Please rename your file or delete the existing one before re-uploading.')
return f"⚠️ A file named '{ppt_file_name}' already exists in the PPT repository. Please rename your file or delete the existing one before re-uploading."
full_ppt_file_id = upload_to_onedrive(file_path, ppt_repo_folder_id,headers)
gr.Info('PPT uploaded to OneDrive..')
full_ppt_file_name = os.path.basename(file_path)
full_ppt_file_path = f"/Projects Apps/PPT Maker/ppt_repo/{full_ppt_file_name}"
# Step 3: Split PPT into individual slides and convert to images
gr.Info('Processing the PPT and indexing ..it may take a while ')
temp_output_folder_slides = "temp_slides"
temp_output_folder_images = "temp_images"
slide_texts = split_and_convert_ppt(file_path, temp_output_folder_slides, temp_output_folder_images)
print('PPT splitted and converted successfully')
# Compile full OCR text
full_text = "\n".join(slide_texts)
gr.Info('AI agent processing the data .')
metadata = generate_metadata_with_retry(full_text, retries=3, max_tokens=5000, decrement=100, model_name="cl100k_base")
# Step 5: Process each slide and prepare metadata for storage
#slides_repo_folder_id = get_folder_id("Projects Apps/PPT Maker/slides_repo")
#slide_image_repo_folder_id = get_folder_id("Projects Apps/PPT Maker/slide_image_repo")
metadata_list = []
gr.Info('Uploading the individual slides and images into repo ')
for i, slide_text in enumerate(slide_texts):
unique_slide_id = f"{file_name}_{ppt_unique_id}_slide_{i + 1}"
slide_file_path = f"{temp_output_folder_slides}/{unique_slide_id}.pptx"
slide_image_path = f"{temp_output_folder_images}/{unique_slide_id}_slide_1.png"
# Upload individual slide (.pptx) to slides_repo
slide_file_id = upload_to_onedrive(slide_file_path, slides_repo_folder_id,headers)
slide_file_path_onedrive = f"/Projects Apps/PPT Maker/slides_repo/{unique_slide_id}.pptx"
print(f'Slide{i} uploaded into Onedrive')
# Upload slide image (.png) to slide_image_repo
thumbnail_file_id = upload_to_onedrive(slide_image_path, slide_image_repo_folder_id,headers)
thumbnail_file_path_onedrive = f"/Projects Apps/PPT Maker/slide_image_repo/{unique_slide_id}.png"
print(f'Image{i} uploaded into Onedrive')
# Generate embedding for the slide
slide_embedding = embedding_model.encode(slide_text).tolist()
short_summary_embedding = embedding_model.encode(metadata.Short_Summary).tolist()
# Prepare metadata for storage
metadata_list.append([
unique_slide_id, # Unique Slide ID
slide_text, # Slide OCR Text
full_text, # PPT OCR Text
str(slide_embedding), # Embedding
str(short_summary_embedding),
ppt_unique_id, # PPT Unique ID
metadata.Suitable_Title, # Suitable Title
metadata.Slide_Category, # Slide Category
metadata.PPT_Owner, # PPT Owner
metadata.Audience_Forum, # Audience Forum
metadata.Short_Summary, # Short Summary
slide_file_path_onedrive, # Slide File Path (.pptx)
slide_file_id, # Slide File ID (.pptx)
full_ppt_file_path, # Full PPT File Path
full_ppt_file_id, # Full PPT File ID
thumbnail_file_path_onedrive, # Thumbnail File Path (.png)
thumbnail_file_id , # Thumbnail File ID (.png)
upload_date # upload date
])
# Clean up temporary files for this slide
os.remove(slide_file_path)
os.remove(slide_image_path)
print('Slides cleared from temp')
# # Clean up temporary folders
# os.rmdir(temp_output_folder_slides)
# os.rmdir(temp_output_folder_images)
# Clean up temporary folders (forcefully deletes all contents inside)
shutil.rmtree(temp_output_folder_slides, ignore_errors=True)
shutil.rmtree(temp_output_folder_images, ignore_errors=True)
print('Temp folders cleared')
gr.Info('Vectorising the meta data and uploading in Onedrive..')
return update_and_upload_metadata_simplified(
metadata_list,
metadata_folder_id,
metadata_with_fulltext_folder_id,
headers
)
except Exception as e:
return f"An error occurred: {str(e)}"
############################################################################### SEARCH PPT ######################################
import requests
from sentence_transformers import SentenceTransformer, CrossEncoder
from sklearn.metrics.pairwise import cosine_similarity
import os
import shutil
import gradio as gr
# Local cache directory for downloaded files
LOCAL_CACHE_DIR = "local_cache"
os.makedirs(LOCAL_CACHE_DIR, exist_ok=True)
# Function to download a file from OneDrive to the local cache
def download_file_from_onedrive(file_path, file_id, headers):
local_file_path = os.path.join(LOCAL_CACHE_DIR, os.path.basename(file_path))
if not os.path.exists(local_file_path): # Avoid re-downloading
download_url = f"https://graph.microsoft.com/v1.0/me/drive/items/{file_id}/content"
response = requests.get(download_url, headers=headers)
if response.status_code != 200:
raise ValueError(f"Failed to download file {file_path}. Error: {response.text}")
with open(local_file_path, "wb") as f:
f.write(response.content)
print(f"βœ… Downloaded: {file_path} -> {local_file_path}")
return local_file_path
# Function to search PPTs
def search_ppts(query, num_results):
global df
gr.Info("Searching the relevant PPTs .")
# Generate query embedding
query_embedding = embedding_model.encode(query).tolist()
# Filter the DataFrame to include only rows where Unique_Slide_ID ends with "slide_1"
df1 = df[df['Unique_Slide_ID'].str.endswith("slide_1", na=False)]
# Compute cosine similarity scores
df1['similarity'] = df1['Short_Summary_Embedding'].apply(
lambda x: cosine_similarity([query_embedding], [eval(x)])[0][0]
)
# Sort by cosine similarity score
df1 = df1.sort_values(by='similarity', ascending=False)
# Get top N results for reranking
top_n = min(50, len(df1)) # Take top 50 results for reranking
top_results = df1.head(top_n)
# Prepare input pairs for cross-encoder reranking
pairs = [(query, row['Short_Summary']) for _, row in top_results.iterrows()]
# Rerank using cross-encoder
gr.Info("Doing Semantic Reranking for most appropriate results ")
rerank_scores = cross_encoder.predict(pairs)
top_results = top_results.copy() # Avoid SettingWithCopyWarning
top_results['rerank_score'] = rerank_scores
# Sort by rerank score
top_results = top_results.sort_values(by='rerank_score', ascending=False)
print(top_results)
# Prepare results
results = []
gr.Info('Downloading PPT images and ppt')
print('Downloading PPT images and ppt')
for _, row in top_results.head(num_results).iterrows():
# Download slide image locally
slide_image_path = download_file_from_onedrive(
row['Thumbnail_File_Path'], row['Thumbnail_File_ID'], headers
)
# Download full PPT locally
ppt_download_link = download_file_from_onedrive(
row['Full_PPT_File_Path'], row['Full_PPT_File_ID'], headers
)
title = row['Suitable_Title']
owner = row['PPT_Owner']
category = row['Slide_Category']
summary = row['Short_Summary']
results.append({
"image": slide_image_path,
"title": title,
"owner": owner,
"category": category,
"summary": summary,
"download_link": ppt_download_link
})
print("downloading complete ")
# Update visibility of rows
visible_rows = min(len(results), num_results)
row_updates = []
row_updates = []
for i in range(20):
if i < len(results):
result = results[i]
row_updates.extend([
gr.update(visible=True), # βœ… Make the row visible
gr.update(value=result["image"], visible=True),
gr.update(value=f"<b>Title:</b> {result['title']}<br><b>Owner:</b> {result['owner']}<br><b>Category:</b> {result['category']}", visible=True),
gr.update(value=result["summary"], visible=True),
gr.update(value=result["download_link"], visible=True),
])
else:
row_updates.extend([gr.update(visible=False)] * 5) # row + 4 components
return row_updates
################################################################ SEARCH SLIDES ########################
import requests
import gradio as gr
import pandas as pd
import tiktoken
import tempfile
from PyPDF2 import PdfReader
from tqdm import tqdm
from pydantic import BaseModel, Field
from phi.agent import Agent, RunResponse
from phi.model.groq import Groq
from sentence_transformers import SentenceTransformer
from sentence_transformers import CrossEncoder
#from gradio_client import Client, handle_file
import os
from pptx import Presentation
from pptx2img import PPTXConverter # For splitting slides
import uuid
import shutil
from PIL import Image
import pandas as pd
import requests
import gradio as gr
from pydantic import BaseModel, Field
from typing import List
import tiktoken
from datetime import datetime
import zipfile
from PIL import Image
import gradio as gr
import threading
import time
# Global variable to store search results
search_results = []
def search_slides(query, num_results):
global search_results # Use the global variable to store results
global df
# # Load metadata file
# gr.Info("Downloading the master file to search..")
# metadata_folder_id = get_folder_id("Projects Apps/PPT Maker/Metadata_file", headers)
# download_metadata_file(metadata_folder_id, headers) # Explicit call to download metadata
# metadata_file = "Master_metadata.csv"
# if not os.path.exists(metadata_file):
# return [gr.update(visible=False) for _ in range(20)], "Metadata file not found."
# df = pd.read_csv(metadata_file)
gr.Info("Searching the relevant slides.")
# Generate query embedding
query_embedding = embedding_model.encode(query).tolist()
# Compute cosine similarity scores
df['similarity'] = df['Slide_Embedding'].apply(
lambda x: cosine_similarity([query_embedding], [eval(x)])[0][0]
)
# Sort by cosine similarity score
df = df.sort_values(by='similarity', ascending=False)
# Get top N results for reranking
top_n = min(50, len(df)) # Take top 50 results for reranking
top_results = df.head(top_n)
# Prepare input pairs for cross-encoder reranking
pairs = [(query, row['Short_Summary']) for _, row in top_results.iterrows()]
# Rerank using cross-encoder
gr.Info("Doing Semantic Reranking for most appropriate results")
rerank_scores = cross_encoder.predict(pairs)
top_results = top_results.copy() # Avoid SettingWithCopyWarning
top_results['rerank_score'] = rerank_scores
# Sort by rerank score
top_results = top_results.sort_values(by='rerank_score', ascending=False)
# Prepare results
results = []
gr.Info('Downloading slide images')
for _, row in top_results.head(num_results).iterrows():
# Download slide image locally
slide_image_path = download_file_from_onedrive(
row['Thumbnail_File_Path'], row['Thumbnail_File_ID'], headers
)
# Download full PPT locally
slide_download_link = download_file_from_onedrive(
row['Slide_File_Path'], row['Slide_File_ID'], headers
)
title = row['Suitable_Title']
owner = row['PPT_Owner']
category = row['Slide_Category']
summary = row['Short_Summary']
results.append({
"image": slide_image_path,
"title": title,
"owner": owner,
"category": category,
"summary": summary,
"slide_path": slide_download_link
})
# Store results in the global variable
search_results = results
# Update visibility of rows
visible_rows = min(len(results), num_results)
row_updates = []
for i in range(20): # Loop through all 20 rows
if i < visible_rows: # For rows with results
result = results[i]
row_updates.extend([
gr.update(visible=True), # Row visibility
gr.update(value=result["image"], visible=True),
gr.update(value=f"<b>Title:</b> {result['title']}<br><b>Owner:</b> {result['owner']}<br><b>Category:</b> {result['category']}", visible=True),
gr.update(value=result["slide_path"], visible=True), # Slide path for identification
gr.update(visible=True) # Checkbox visibility
])
else: # For rows without results
row_updates.extend([gr.update(visible=False)] * 6) # Row + 5 components
return row_updates
def combine_slides_as_zip(*checkbox_values):
"""
Collects selected individual slide files and zips them.
Returns the path to the ZIP file.
"""
selected_files = [
result["slide_path"] for result, selected in zip(search_results, checkbox_values) if selected
]
if not selected_files:
return "No slides selected."
zip_filename = os.path.join(LOCAL_CACHE_DIR, "selected_slides.zip")
with zipfile.ZipFile(zip_filename, 'w') as zipf:
for file_path in selected_files:
arcname = os.path.basename(file_path) # Only filename in zip
zipf.write(file_path, arcname=arcname)
return zip_filename
# Background thread to wait for login
def background_login(flow):
global headers
result = app.acquire_token_by_device_flow(flow)
access_token = result["access_token"]
if "access_token" in result:
access_token_state["token"] = result["access_token"]
access_token = result["access_token"]
headers = {
"Authorization": f"Bearer {access_token}",
"Content-Type": "application/json"
}
else:
access_token_state["token"] = "ERROR"
def login_action():
flow = app.initiate_device_flow(scopes=SCOPES)
flow_state["flow"] = flow
login_url = flow["verification_uri"]
login_code = flow["user_code"]
instructions = f"""
<p style='text-align:center; color:#1E3A8A;'>Please go to the following link to authenticate:</p>
<p style='text-align:center;'><a href='{login_url}' target='_blank'>{login_url}</a></p>
<p style='text-align:center;'>Enter the code: <strong>{login_code}</strong></p>
"""
# Start background login thread
threading.Thread(target=background_login, args=(flow,), daemon=True).start()
return gr.update(value=instructions, visible=True)
# Check token and control UI switch
def check_login_status():
token = access_token_state["token"]
if token == "ERROR":
return gr.update(visible=True, value="❌ Login failed.Click Login button again to Try again"), gr.update(visible=True), gr.update(visible=False)
elif token:
return gr.update(value="", visible=False), gr.update(visible=False), gr.update(visible=True)
else:
return gr.update(visible=False), gr.update(visible=True), gr.update(visible=False)
def validate_admin_access(username, password):
if username == ADMIN_USERNAME and password == ADMIN_PASSWORD:
return (
gr.update(visible=False), # Hide admin login form
gr.update(visible=True), # Show admin upload UI
gr.update(visible=False, value="") # Clear any error
)
else:
return (
gr.update(visible=True),
gr.update(visible=False),
gr.update(visible=True, value="❌ Invalid credentials")
)
def load_and_store_metadata_df():
global temp_file_path
# Load metadata file
gr.Info("Downloading the master file ..We will be ready shortly")
metadata_folder_id = get_folder_id("Projects Apps/PPT Maker/Metadata_file", headers)
temp_file_path =download_metadata_file(metadata_folder_id, headers) # Explicit call to download metadata
# metadata_file = "Master_metadata.csv"
# temp_file_path = os.path.join("/tmp", metadata_file_name)
if not os.path.exists(temp_file_path):
return [gr.update(visible=False) for _ in range(20)], "Metadata file not found."
# if not os.path.exists(metadata_file):
# return [gr.update(visible=False) for _ in range(20)], "Metadata file not found."
#CSS for checkboxes
css="""
.gr-button {
background-color: #1E3A8A;
color: white;
}
/* Style for checkbox column */
.checkbox-column {
background-color: #EFF6FF;
border-radius: 10px;
padding: 10px;
margin-top: 8px;
margin-bottom: 8px;
box-shadow: 0 1px 4px rgba(0,0,0,0.1);
transition: box-shadow 0.3s ease;
}
.checkbox-column:hover {
box-shadow: 0 2px 8px rgba(0,0,0,0.2);
}
/* Style the checkbox directly */
.gr-checkbox {
font-weight: bold;
color: #1D4ED8;
}
"""
# # # MAIN APP # # #
with gr.Blocks(css=css) as demo:
with gr.Column(visible=True) as login_section:
gr.HTML("<h1 style='text-align:center; color:#1E3A8A;'>NCTC SlideFinder</h1>")
# 🌟 Subheading
gr.HTML("<h3 style='text-align:center; color:#0F766E;'>PPT Repo and Smart Search Powered by AI</h3>")
gr.HTML("""
<div style='text-align:center;'>
<img src='/file=logo.jpg' width='200' height='200' style='margin-top:10px;' />
</div>
""")
login_button = gr.Button("πŸ” Login")
auth_instructions = gr.HTML(visible=False)
login_error = gr.Textbox(visible=False, interactive=False, label="", show_label=False)
status_checker = gr.Button("βœ… Check Login Status")
with gr.Column(visible=False) as main_app_section:
gr.Markdown("<h2 style='text-align:center; color:#0F766E;'>Welcome to NCTC PPT Repository</h2>")
with gr.Tab("πŸ“Š Stats Dashboard"):
with gr.Column() as dashboard_section:
gr.Markdown("### πŸ“Š Dashboard Overview")
with gr.Row():
total_ppt_box = gr.HTML()
total_slides_box = gr.HTML()
with gr.Row():
chart_output = gr.BarPlot(x="Month", y="PPT Uploads", label="Monthly PPT Uploads")
latest_ppts_output = gr.HTML()
with gr.Tab("Upload PPT"):
# file_input = gr.File(label="Upload PPT File")
# output_text = gr.Textbox(label="Processing Status")
# submit_button = gr.Button("Process")
# submit_button.click(process_presentation, inputs=file_input, outputs=output_text)
with gr.Column() as admin_access_section:
gr.Markdown("### πŸ” Admin Access Required")
username_input = gr.Textbox(label="Username", placeholder="Enter username")
password_input = gr.Textbox(label="Password", type="password", placeholder="Enter password")
admin_login_msg = gr.Textbox(visible=False, interactive=False, show_label=False)
admin_login_button = gr.Button("πŸ”“ Proceed")
with gr.Column(visible=False) as admin_upload_ui:
file_input = gr.File(label="Upload PPT File")
output_text = gr.Textbox(label="Processing Status")
submit_button = gr.Button("Process")
submit_button.click(process_presentation, inputs=file_input, outputs=output_text)
admin_login_button.click(
validate_admin_access,
inputs=[username_input, password_input],
outputs=[admin_access_section, admin_upload_ui, admin_login_msg]
)
with gr.Tab("Search PPT"):
query_input = gr.Textbox(label="Enter Search Query", placeholder="e.g., Risk Management")
num_results_input = gr.Number(label="Number of Results", value=5, minimum=1, maximum=20)
search_button = gr.Button("πŸ” Search")
result_rows = []
result_components = []
for i in range(20):
with gr.Row(visible=False) as row:
with gr.Column(scale=2): # image small
image_output = gr.Image(label="Slide Image")
with gr.Column(scale=1): # image small
info_output = gr.HTML(label="PPT Info")
with gr.Column(scale=2): # image small
summary_output = gr.Textbox(label="Short Summary", lines=3)
with gr.Column(scale=1): # image small
# download_button = gr.Button("Download PPT")
download_file = gr.File( label="πŸ“₯ Download PPT")
result_rows.append(row) # βœ… Track rows
result_components.extend([row, image_output, info_output, summary_output, download_file])
search_button.click(
search_ppts,
inputs=[query_input, num_results_input],
outputs=result_components
)
with gr.Tab("Search and Combine Slides"):
query_input = gr.Textbox(label="Enter Search Query to search slides", placeholder="e.g., Risk Management")
num_results_input = gr.Number(label="Number of Slides you need", value=5, minimum=1, maximum=20)
search_button = gr.Button("πŸ” Search")
result_rows = []
result_components = []
checkboxes = []
for i in range(20):
with gr.Row(visible=False) as row:
with gr.Column(scale=4): # Image small
image_output = gr.Image(label="Slide Image")
with gr.Column(scale=2): # Info small
info_output = gr.HTML(label="Slide Info")
# with gr.Column(scale=2): # Summary small
# summary_output = gr.Textbox(label="Short Summary", lines=3)
with gr.Column(scale=1): # Slide ID small
download_file = gr.File( label="πŸ“₯ Download Slide")
#slide_id_output = gr.Textbox(label="Slide ID", interactive=False)
with gr.Column(scale=1, elem_classes=["checkbox-column"]): # Checkbox small
checkbox = gr.Checkbox(label="Select to Combine")
checkboxes.append(checkbox)
result_rows.append(row) # Track rows
result_components.extend([row, image_output, info_output, download_file, checkbox])
combine_button = gr.Button("Combine Selected Slides")
combined_ppt_output = gr.File(label="Download Combined PPT")
search_button.click(
search_slides,
inputs=[query_input, num_results_input],
outputs=result_components
)
combine_button.click(
combine_slides_as_zip,
inputs=checkboxes,
outputs=gr.File(label="Download ZIP")
)
login_button.click(login_action, inputs=[], outputs=[auth_instructions])
status_checker.click(
check_login_status,
inputs=[],
outputs=[login_error, login_section, main_app_section]
).then(
fn=load_and_store_metadata_df,
inputs=[],
outputs=[]
).then(
fn=update_dashboard,
inputs=[],
outputs=[dashboard_section, total_ppt_box, total_slides_box, chart_output, latest_ppts_output]
)
demo.launch(debug=True, allowed_paths=[LOCAL_CACHE_DIR])