Update app.py

app.py

@@ -22,33 +22,44 @@ app.add_middleware(
 # Mount static files for Hugging Face Spaces
 app.mount("/", StaticFiles(directory="static", html=True), name="static")
 
-# Load data
-faq_df = pd.read_csv("BankFAQs.csv", usecols=["Question", "Answer"])
-questions = faq_df["Question"].astype(str).tolist()
-answers = faq_df["Answer"].astype(str).tolist()
+# Global variables for lazy initialization
+model = None
+chunk_index = None
+chunked_questions = None
+chunked_answers = None
 
-# Chunking function: split text into sentences
-sentence_splitter = re.compile(r'(?<=[.!?]) +')
-def chunk_text(text):
-    return [chunk.strip() for chunk in sentence_splitter.split(text) if chunk.strip()]
+def initialize_model():
+    """Initialize the model and data on first use"""
+    global model, chunk_index, chunked_questions, chunked_answers
+    
+    if model is None:
+        # Load data
+        faq_df = pd.read_csv("BankFAQs.csv", usecols=["Question", "Answer"])
+        questions = faq_df["Question"].astype(str).tolist()
+        answers = faq_df["Answer"].astype(str).tolist()
 
-# Prepare chunked data
-chunked_questions = []  # Parent question for each chunk
-chunks = []             # The actual chunk text
-chunked_answers = []    # Full answer for reference
-for q, a in zip(questions, answers):
-    answer_chunks = chunk_text(a)
-    for chunk in answer_chunks:
-        chunked_questions.append(q)
-        chunks.append(chunk)
-        chunked_answers.append(a)
+        # Chunking function: split text into sentences
+        sentence_splitter = re.compile(r'(?<=[.!?]) +')
+        def chunk_text(text):
+            return [chunk.strip() for chunk in sentence_splitter.split(text) if chunk.strip()]
 
-# Load model and build index
-model = SentenceTransformer("all-MiniLM-L6-v2")
-chunk_embeddings = model.encode(chunks)
-chunk_embeddings = np.array(chunk_embeddings).astype("float32")
-chunk_index = faiss.IndexFlatL2(chunk_embeddings.shape[1])
-chunk_index.add(chunk_embeddings)
+        # Prepare chunked data
+        chunked_questions = []  # Parent question for each chunk
+        chunks = []             # The actual chunk text
+        chunked_answers = []    # Full answer for reference
+        for q, a in zip(questions, answers):
+            answer_chunks = chunk_text(a)
+            for chunk in answer_chunks:
+                chunked_questions.append(q)
+                chunks.append(chunk)
+                chunked_answers.append(a)
+
+        # Load model and build index
+        model = SentenceTransformer("all-MiniLM-L6-v2")
+        chunk_embeddings = model.encode(chunks)
+        chunk_embeddings = np.array(chunk_embeddings).astype("float32")
+        chunk_index = faiss.IndexFlatL2(chunk_embeddings.shape[1])
+        chunk_index.add(chunk_embeddings)
 
 class QueryRequest(BaseModel):
     query: str
@@ -56,6 +67,9 @@ class QueryRequest(BaseModel):
 
 @app.post("/search")
 async def search_faq(req: QueryRequest):
+    # Initialize model on first request
+    initialize_model()
+    
     query_embedding = model.encode([req.query]).astype("float32")
     D, I = chunk_index.search(query_embedding, req.k)
     # Calculate cosine similarity from L2 distance