Update app.py

app.py

@@ -208,9 +208,28 @@ def query_crm_data_with_context(prompt, top_k=3):
 
 
 
+
+storage_path = r"C:\Users\bhagy\OneDrive\Desktop\INFOSYS PROJECT\chromadb_storage"  # Update with your storage path
+
+# Ensure the storage path exists
+if not os.path.exists(storage_path):
+    print(f"Storage path {storage_path} does not exist. Creating it...")
+    os.makedirs(storage_path)
+
+# Clear any corrupted data
+if os.path.exists(storage_path):
+    print(f"Removing existing storage at {storage_path}...")
+    shutil.rmtree(storage_path)
+    os.makedirs(storage_path)
+
+# Initialize the sentence transformer model
 sentence_model = SentenceTransformer('sentence-transformers/all-MiniLM-L6-v2')
-faiss_index = faiss.IndexFlatL2(384)
 
+# Create Faiss index with the correct dimensionality (make sure this matches the embedding size)
+expected_dim = 384  # Example value for sentence embeddings
+faiss_index = faiss.IndexFlatL2(expected_dim)
+
+# Load objection responses
 def load_objection_responses(csv_file_path):
     try:
         df = pd.read_csv(csv_file_path)
@@ -220,43 +239,23 @@ def load_objection_responses(csv_file_path):
         print(f"Error loading objections CSV: {e}")
         return {}
 
+# Load objections and encode them into embeddings
 objection_response_pairs = load_objection_responses(r"C:\Users\bhagy\OneDrive\Desktop\INFOSYS PROJECT\objections_responses.csv")
 objections = list(objection_response_pairs.keys())
 objection_embeddings = sentence_model.encode(objections)
 objection_embeddings = np.array(objection_embeddings, dtype="float32")
-print(f"Shape of objection_embeddings: {objection_embeddings.shape}")
 
-# Assuming you know the expected dimension of the embeddings (e.g., 768)
-expected_dim = 384  # Example value for sentence embeddings (replace with the actual dimension)
-if len(objection_embeddings.shape) == 1:
-    objection_embeddings = objection_embeddings.reshape(-1, 1)
+# Check the shape of the embeddings
+print(f"Shape of objection_embeddings: {objection_embeddings.shape}")
 
 # Check if the embeddings dimensionality matches the expected dimension
 if objection_embeddings.shape[1] != expected_dim:
     raise ValueError(f"Dimensionality of embeddings {objection_embeddings.shape[1]} does not match expected dimension {expected_dim}.")
 
-# If the shape is (1, d) (e.g., a single embedding), reshape it to (1, d)
-if len(objection_embeddings.shape) == 1:
-    objection_embeddings = objection_embeddings.reshape(1, -1)  # Reshape for a single embedding
-elif len(objection_embeddings.shape) == 2:
-    pass  # The shape is already in the expected form (num_samples, dim)
-else:
-    raise ValueError(f"Unexpected shape for objection embeddings: {objection_embeddings.shape}")
-
-# Create Faiss index with the correct dimensionality (make sure this matches the embedding size)
-faiss_index = faiss.IndexFlatL2(expected_dim)
-
-# Now add the embeddings to the Faiss index
+# Add the embeddings to the Faiss index
 faiss_index.add(objection_embeddings)
 
 print(f"Successfully added {objection_embeddings.shape[0]} embeddings to the Faiss index.")
-# if len(objection_embeddings.shape) == 1:
-#     objection_embeddings = objection_embeddings.reshape(1, -1)  # Reshape for a single embedding
-# elif len(objection_embeddings.shape) == 2:
-#     pass  
-# else:
-#     raise ValueError("Unexpected shape for objection embeddings:", objection_embeddings.shape)
-# faiss_index.add(objection_embeddings)
 
 def find_closest_objection(query):
     query_embedding = sentence_model.encode([query])