Update app.py

app.py

@@ -66,9 +66,24 @@ def load_document(file: NamedTemporaryFile, parser: str = "llamaparse") -> List[
     else:
         raise ValueError("Invalid parser specified. Use 'pypdf' or 'llamaparse'.")
 
+# Function to get the embeddings model
 def get_embeddings():
     return HuggingFaceEmbeddings(model_name="avsolatorio/GIST-Embedding-v0")
 
+# Function to encode text with specific financial focus
+def get_embedding(text):
+    instruction = """
+    Encode this text with a focus on financial information. 
+    Pay special attention to:
+    1. Numerical data related to earnings, revenue, and other financial metrics
+    2. Statements about company performance and future outlook
+    3. Comparisons to previous periods or industry benchmarks
+    4. Key financial terms and their context
+    Ignore general boilerplate text and focus on the substantive financial content.
+    """
+    embedding_model = get_embeddings()
+    return embedding_model.encode(instruction + "\n\n" + text)
+
 # Add this at the beginning of your script, after imports
 DOCUMENTS_FILE = "uploaded_documents.json"
 
@@ -94,7 +109,7 @@ def update_vectors(files, parser):
         logging.warning("No files provided for update_vectors")
         return "Please upload at least one PDF file.", display_documents()
     
-    embed = get_embeddings()
+    embed = get_embedding()
     total_chunks = 0
     
     all_data = []
@@ -148,7 +163,7 @@ def delete_documents(selected_docs):
     if not selected_docs:
         return "No documents selected for deletion.", display_documents()
     
-    embed = get_embeddings()
+    embed = get_embedding()
     database = FAISS.load_local("faiss_database", embed, allow_dangerous_deserialization=True)
     
     deleted_docs = []
@@ -322,7 +337,7 @@ def respond(message, history, model, temperature, num_calls, use_web_search, sel
 #                logging.info(f"Generated Response (first line): {first_line}")
                 yield response
         else:
-            embed = get_embeddings()
+            embed = get_embedding()
             if os.path.exists("faiss_database"):
                 database = FAISS.load_local("faiss_database", embed, allow_dangerous_deserialization=True)
                 retriever = database.as_retriever(search_kwargs={"k": 20})
@@ -453,7 +468,7 @@ After writing the document, please provide a list of sources used in your respon
 def get_response_from_pdf(query, model, selected_docs, num_calls=3, temperature=0.2):
     logging.info(f"Entering get_response_from_pdf with query: {query}, model: {model}, selected_docs: {selected_docs}")
     
-    embed = get_embeddings()
+    embed = get_embedding()
     if os.path.exists("faiss_database"):
         logging.info("Loading FAISS database")
         database = FAISS.load_local("faiss_database", embed, allow_dangerous_deserialization=True)