Update pipeline.py

pipeline.py

@@ -8,8 +8,8 @@ from langchain.embeddings import HuggingFaceEmbeddings
 from langchain.vectorstores import FAISS
 from langchain.chains import RetrievalQA
 from smolagents import CodeAgent, DuckDuckGoSearchTool, ManagedAgent, LiteLLMModel
-from langchain.llms.base import LLM 
 import subprocess  # Import subprocess to run shell commands
+from langchain.llms.base import LLM  # Import LLM
 
 # Import the chain builders from our separate files
 from classification_chain import get_classification_chain
@@ -38,6 +38,8 @@ def install_spacy_model():
 # Call the function to install the spaCy model if needed
 install_spacy_model()
 
+# Load the spaCy model globally
+nlp = spacy.load("en_core_web_sm")
 
 # Function to extract the main topic using NER
 def extract_main_topic(query: str) -> str:
@@ -45,7 +47,7 @@ def extract_main_topic(query: str) -> str:
     Extracts the main topic from the user's query using spaCy's NER.
     Returns the first named entity or noun found in the query.
     """
-    doc = nlp(query)
+    doc = nlp(query)  # Use the globally loaded spaCy model
     
     # Try to extract the main topic as a named entity (person, product, etc.)
     main_topic = None
@@ -100,9 +102,11 @@ def build_rag_chain(llm_model: LiteLLMModel, vectorstore: FAISS) -> RetrievalQA:
         def _call(self, prompt: str, stop: Optional[list] = None, **kwargs) -> str:
             messages = [{"role": "user", "content": prompt}]
             return llm_model(messages, stop_sequences=stop)
+
         @property
         def _llm_type(self) -> str:
             return "custom_gemini"
+    
     retriever = vectorstore.as_retriever(search_type="similarity", search_kwargs={"k": 3})
     gemini_as_llm = GeminiLangChainLLM()
     rag_chain = RetrievalQA.from_chain_type(