Update app.py

app.py

@@ -38,6 +38,8 @@ def flatten_metadata(metadata):
     return str(metadata)  # If it's not a dict, just return the string version
 
 def metadata_func(record, additional_fields=None):
+    is_winner = record.get("Ranking", "").lower() == "winner"
+    
     return {
         "title": record.get("Title", ""),
         "organization": record.get("Organization", ""),
@@ -59,9 +61,9 @@ def metadata_func(record, additional_fields=None):
             "facebook": record.get("FaceBook", "")
         }),
         "working_area": record.get("Working Areas in LA", ""),
-        "zipcode": record.get("Zipcode", "")
+        "zipcode": record.get("Zipcode", ""),
+        "priority": 1 if is_winner else 0  # Assign a priority value
     }
-
 # Load the JSON data with custom metadata and content key
 loader = JSONLoader(
     file_path='data.json',
@@ -78,10 +80,15 @@ data = loader.load()
 from langchain_text_splitters import RecursiveCharacterTextSplitter
 text_splitter = RecursiveCharacterTextSplitter(
     chunk_size=760,  
-    chunk_overlap=100,  
+    chunk_overlap=50,  
     add_start_index=True
 )
 all_splits = text_splitter.split_documents(data)
+priority_docs = [doc for doc in all_splits if doc.metadata.get("priority", 0) == 1]
+other_docs = [doc for doc in all_splits if doc.metadata.get("priority", 0) == 0]
+
+# Combine priority documents with other documents, ensuring winners from 2023 are prioritized
+all_docs = priority_docs + other_docs
 
 # -------------------------------
 # Set Up Retrievers
@@ -97,7 +104,7 @@ if os.path.exists(persist_directory) and os.listdir(persist_directory):
     print("Loaded vector store from persist directory.")
 else:
     vectorstore = Chroma.from_documents(
-        documents=all_splits, 
+        documents=all_docs, 
         embedding=OpenAIEmbeddings(), 
         persist_directory=persist_directory
     )
@@ -118,7 +125,6 @@ retriever = ensemble_retriever
 # -------------------------------
 
 system_prompt = (
-
     "You are the LA2050 Navigator, an AI-powered chatbot designed to help users explore organizations and community initiatives within the Goldhirsh Foundation’s LA2050 Ideas Hub. "
 
     "Your role is to provide concise, personalized recommendations, guide users toward supporting these organizations and initiatives, and answer relevant questions about the Goldhirsh Foundation, LA2050, and its projects. "
@@ -129,6 +135,8 @@ system_prompt = (
 
     "Prioritize nonprofit organizations awarded by the Goldhirsh Foundation (designated 'winner') and those with multiple proposal submissions. "
 
+    "If the user asks about the LA2050 grant winners for a specific year, be sure to reference the 'Year' and 'LA 2050 Grant Status' fields in your responses. If the organization was awarded the grant in that year, mention it explicitly."
+
     "Use the data files as your primary source of information. If information is unavailable, acknowledge it and guide the user to relevant resources. "
 
     "Maintain a polite, helpful, respectful, and enthusiastic tone at all times. "
@@ -136,11 +144,11 @@ system_prompt = (
     "If the user responds with a follow-up confirmation (e.g. 'yes') after a previous answer, please expand on that topic with additional information. "
 
     "\n\n{context}"
-
 )
 
 
 
+
 prompt = ChatPromptTemplate.from_messages(
     [
         ("system", system_prompt),