Create app.py

app.py

@@ -0,0 +1,107 @@
+import pandas as pd
+import requests
+from bs4 import BeautifulSoup
+import re
+from sentence_transformers import SentenceTransformer, util
+from huggingface_hub import InferenceClient
+import gradio as gr
+import os
+
+print("1. Ingesting Unstructured Data...")
+# We use a robust fallback block of the script to ensure the app never crashes for a recruiter
+fallback_script = """
+[ Scene starts in FBI facility ]
+RESSLER: He’s demanding to speak with you.
+ELIZABETH: Me? Why me? I don't know him.
+RESSLER: We don't know why. But we need to hear what he has to say.
+[ Elizabeth walks into the interrogation room ]
+REDDINGTON: You got rid of your highlights. You look much less... Baltimore.
+ELIZABETH: You asked to see me. Why?
+REDDINGTON: I'm a criminal, Lizzie. You're a profiler. You tell me. 
+ELIZABETH: I think you're bored. I think you're using us.
+REDDINGTON: I'm going to make you famous. I have a list. A blacklist of the worst criminals in the world. And I will give them to you.
+REDDINGTON: Starting with Ranko Zamani. He is in town, Lizzie. And he plans to kidnap the general's daughter.
+"""
+
+url = "https://subslikescript.com/series/The_Blacklist-2741602/season-1/episode-1-Pilot"
+headers = {'User-Agent': 'Mozilla/5.0'}
+
+try:
+    response = requests.get(url, headers=headers, timeout=5)
+    response.raise_for_status()
+    soup = BeautifulSoup(response.text, 'html.parser')
+    script_div = soup.find('div', class_='full-script')
+    raw_text = script_div.get_text(separator='\n') if script_div else fallback_script
+    print("Scraped live data successfully.")
+except Exception:
+    print("Live scrape blocked by server. Using fallback script.")
+    raw_text = fallback_script
+
+print("2. Structuring and Chunking the Data...")
+lines = raw_text.split('\n')
+structured_data = []
+dialogue_pattern = re.compile(r"^([A-Z\s]+):\s*(.*)")
+
+for line in lines:
+    line = line.strip()
+    if not line: continue
+    match = dialogue_pattern.match(line)
+    if match:
+        structured_data.append({"Character": match.group(1).strip(), "Dialogue": match.group(2).strip()})
+
+df = pd.DataFrame(structured_data)
+df['Full_Line'] = df['Character'] + ": " + df['Dialogue']
+
+# Sliding Window Context (3 lines total per chunk)
+window_size = 1
+context_chunks = []
+for i in range(len(df)):
+    start = max(0, i - window_size)
+    end = min(len(df), i + window_size + 1)
+    chunk = "\n".join(df['Full_Line'].iloc[start:end])
+    context_chunks.append({"context_block": chunk})
+
+ai_df = pd.DataFrame(context_chunks)
+
+print("3. Vectorizing Data & Connecting LLM...")
+model = SentenceTransformer('all-MiniLM-L6-v2')
+vector_database = model.encode(ai_df['context_block'].tolist())
+
+hf_token = os.environ.get("HF_TOKEN")
+client = InferenceClient(model="meta-llama/Meta-Llama-3-8B-Instruct", token=hf_token)
+
+def search_script(user_query):
+    query_vector = model.encode(user_query)
+    hits = util.semantic_search(query_vector, vector_database, top_k=2)
+    
+    retrieved_contexts = [ai_df['context_block'].iloc[hit['corpus_id']] for hit in hits[0]]
+    combined_context = "\n---\n".join(retrieved_contexts)
+    
+    messages = [
+        {"role": "system", "content": "You are a data analyst answering questions using ONLY the provided script excerpts. Do not invent lore."},
+        {"role": "user", "content": f"Question: {user_query}\n\nScript Excerpts:\n{combined_context}"}
+    ]
+    
+    try:
+        response = client.chat.completions.create(messages=messages, max_tokens=250, temperature=0.2)
+        return response.choices[0].message.content.strip(), combined_context
+    except Exception as e:
+        return f"Error: {e}", combined_context
+
+print("4. Launching Interface...")
+with gr.Blocks() as demo:
+    gr.Markdown("# 🎬 Unstructured Data AI: TV Script Search")
+    gr.Markdown("This agent processes messy, unstructured dialogue. Ask a question about the plot, and the AI will search the script to find the exact context.")
+    
+    with gr.Row():
+        with gr.Column():
+            input_query = gr.Textbox(lines=3, label="Ask a question (e.g., 'Who is Ranko Zamani?' or 'Why does Reddington want Elizabeth?')")
+            submit_btn = gr.Button("Search Script")
+        
+        with gr.Column():
+            output_reply = gr.Textbox(lines=5, label="AI Answer")
+            retrieved_doc = gr.Textbox(lines=5, label="Raw Context Retrieved by Python")
+
+    submit_btn.click(fn=search_script, inputs=input_query, outputs=[output_reply, retrieved_doc])
+
+demo.launch()