Update app.py

added feature to query an LLM about the top selection

app.py

@@ -6,6 +6,7 @@ import numpy as np
 from sentence_transformers import SentenceTransformer, CrossEncoder
 from huggingface_hub import HfApi, hf_hub_download
 from huggingface_hub.utils import EntryNotFoundError, RepositoryNotFoundError
+from huggingface_hub import InferenceClient
 import pypdf
 import docx
 import time
@@ -91,6 +92,38 @@ def recursive_chunking(text, source, chunk_size=500, overlap=100):
             chunks.append({"text": chunk_text, "source": source})
     return chunks
 
+def ask_llm(query, context):
+    """
+    Sends the user query and the retrieved document text to a free, hosted LLM.
+    """
+    if not HF_TOKEN:
+        return "Error: HF_TOKEN is missing. Cannot contact AI."
+    
+    # We use Mistral-7B-Instruct because it is fast, follows instructions well, 
+    # and is usually available on the free tier.
+    repo_id = "mistralai/Mistral-7B-Instruct-v0.3"
+    
+    client = InferenceClient(model=repo_id, token=HF_TOKEN)
+    
+    prompt = f"""
+    You are a Senior Navy Yeoman and Subject Matter Expert. 
+    Analyze the following Navy document excerpt and answer the user's question based ONLY on that text.
+    
+    USER QUESTION: "{query}"
+    
+    DOCUMENT EXCERPT:
+    "{context}"
+    
+    Your Answer (Be concise, professional, and cite the document):
+    """
+    
+    try:
+        # stream=True makes it look cool (typewriter effect) but standard return is easier for now
+        response = client.text_generation(prompt, max_new_tokens=400)
+        return response
+    except Exception as e:
+        return f"AI Error: {e}"
+
 # --- CORE SEARCH ENGINE ---
 class DocSearchEngine:
     def __init__(self):