changes - logging

.gitignore

@@ -1 +1,2 @@
-.venv
+.venv
+.env

app.py

@@ -15,8 +15,13 @@ from google.oauth2 import service_account
 import os
 from langchain_groq import ChatGroq
 import groq
+from datetime import datetime
 
-sheet_url = os.getenv("SHEET_URL")
+# Load environment variables
+from dotenv import load_dotenv
+load_dotenv()
+
+SHEET_URL = os.getenv("SHEET_URL")
 GOOGLESHEETS_CREDENTIALS = os.getenv("GOOGLESHEETS_CREDENTIALS")
 GROQ_API_KEY = os.environ.get("GROQ_API_KEY")
 MODEL = "meta-llama/llama-4-scout-17b-16e-instruct"
@@ -24,11 +29,11 @@ MODEL = "meta-llama/llama-4-scout-17b-16e-instruct"
 class TransactionParser(BaseModel):
     """This Pydantic class is used to parse the transaction message."""
 
-    amount: str = Field(description="The amount of the transaction strictly in decimal format. If the transaction is a credit or a reversal, then include negative sign. Otherwise it should always be positive. DO not insert currency symbol.", example="123.45")
+    amount: str = Field(description="The amount of the transaction strictly in decimal format. Do not insert currency symbol.", example="123.45")
     dr_or_cr: str = Field(description="Identify if the transaction was debit (spent) or credit (received). Strictly choose one of the values - Debit or Credit")
-    receiver: str = Field(description="The recipient of the transaction. Identify the Merchant Name from the value.")
+    receiver: str = Field(description="The recipient of the transaction. Identify the Merchant Name from the message text.")
     category: str = Field(description="The category of the transaction. The category of the transaction is linked to the Merchant Name. Strictly choose from one the of values - Shopping,EMI,Education,Miscellaneous,Grocery,Utility,House Help,Travel,Transport")
-    transaction_date: str = Field(description="The date of the transaction strictly in yyyy-mm-dd format. If the year is not provided then use current year.")
+    transaction_date: str = Field(description="Use current system date strictly in yyyy-mm-dd format.")
     transaction_origin: str = Field(description="The origin of the transaction. Provide the card or account number as well.")
 
 class AgentState(TypedDict):
@@ -38,8 +43,8 @@ class Agent:
     def __init__(self, model, system=""):
         self.system = system
         graph = StateGraph(AgentState)
-        graph.add_node("classify_txn_type", self.classify_txn_type, retry=RetryPolicy(retry_on=[groq.BadRequestError], max_attempts=2))
-        graph.add_node("parse_message", self.parse_message, retry=RetryPolicy(retry_on=[groq.BadRequestError], max_attempts=2))
+        graph.add_node("classify_txn_type", self.classify_txn_type, retry=RetryPolicy(retry_on=[groq.APIConnectionError], max_attempts=5))
+        graph.add_node("parse_message", self.parse_message, retry=RetryPolicy(retry_on=[groq.APIConnectionError], max_attempts=5))
         graph.add_node("write_message", self.write_message)
         graph.add_conditional_edges(
             "classify_txn_type",
@@ -53,17 +58,17 @@ class Agent:
         self.model = model
 
     def classify_txn_type(self, state: AgentState) -> AgentState:
-        print("Classifying transaction type...")
+        print(f"{datetime.now()}: Classifying transaction type...")
         messages = state["messages"]
         if self.system:
             messages = [SystemMessage(content=self.system)] + messages
 
         message = self.model.invoke(messages)
-        print("Classifying transaction type completed.")
+        print(f"{datetime.now()}: Classifying transaction type completed.")
         return {"messages": [message]}
     
     def parse_message(self, state: AgentState) -> AgentState:
-        print("Parsing transaction message...")
+        print(f"{datetime.now()}: Parsing transaction message...")
         message = state["messages"][0]#.content
         system = """
         You are a helpful assistant skilled at parsing transaction messages and providing structured responses.
@@ -73,17 +78,18 @@ class Agent:
         prompt = ChatPromptTemplate.from_messages([("system", system), ("human", human)])
         chain = prompt | self.model.with_structured_output(TransactionParser)
         result = chain.invoke({"topic": message})
-        print("Parsing transaction message completed.")        
+        print(f"{datetime.now()}: Parsing transaction message completed.")        
         return {"messages": [result]}
 
     def write_message(self, state: AgentState) -> AgentState:
-        print("Writing transaction message to Google Sheets...")
+        print(f"{datetime.now()}: Writing transaction message to Google Sheets...")
         result = state["messages"][-1]
         SCOPES = ('https://www.googleapis.com/auth/spreadsheets', 'https://www.googleapis.com/auth/drive')
+
         service_account_info = json.loads(GOOGLESHEETS_CREDENTIALS)
-        my_credentials = service_account.Credentials.from_service_account_info(service_account_info, scopes=SCOPES)
-        client = pygsheets.authorize(custom_credentials=my_credentials)
-        worksheet = client.open_by_url(sheet_url)
+        credentials = service_account.Credentials.from_service_account_info(service_account_info, scopes=SCOPES)
+        client = pygsheets.authorize(custom_credentials=credentials)
+        worksheet = client.open_by_url(SHEET_URL)
         wk = worksheet[0]
         # Get number of rows in the worksheet
         df = wk.get_as_df(start='A1', end='G999')
@@ -95,7 +101,7 @@ class Agent:
         wk.update_value(f'E{nrows+2}', result.transaction_date)
         wk.update_value(f'F{nrows+2}', result.transaction_origin)
         wk.update_value(f'G{nrows+2}', state["messages"][0])
-        print("Writing transaction message to Google Sheets completed.")
+        print(f"{datetime.now()}: Writing transaction message to Google Sheets completed.")
         return {"messages": ["Transaction Completed"]}
         
     def check_txn_and_decide(self, state: AgentState):
@@ -109,7 +115,8 @@ class Agent:
 prompt = """You are a smart assistant adept at classifying different messages. \
 You will be penalized heavily for incorrect classification. \
 Your task is to classify the message into one of the following categories: \
-Transaction, OTP, Promotional, Scheduled. \
+Transaction, OTP, Promotional, Scheduled, Non-Transaction. \
+Consider Salary Credit, Interest Credit, Redemtions as non-transactional messages.
 Output the classification in a structured format like below. \
 {"classification": "OTP"} \
 """

requirements.txt

@@ -4,4 +4,5 @@ langchain-ollama
 langgraph
 pygsheets
 pandas
-langchain-groq
+langchain-groq
+dotenv