push

gemini_langchain_agent.py

@@ -3,7 +3,8 @@ import uuid
 from langchain_google_genai import ChatGoogleGenerativeAI
 import pandas as pd
 from langchain_core.prompts import ChatPromptTemplate
-from langchain_experimental.tools import PythonAstREPLTool
+# Removed the import for PythonAstREPLTool as it causes a duplicate tool declaration
+# from langchain_experimental.tools import PythonAstREPLTool 
 from langchain_experimental.agents import create_pandas_dataframe_agent
 from dotenv import load_dotenv
 import numpy as np
@@ -26,15 +27,18 @@ llm_instances = [
 ]
 current_instance_index = 0  # Track current instance being used
 
-def create_agent(llm, data, tools):
-    """Create agent with tool names"""
+# Modified create_agent function: Removed the 'tools' parameter and 'extra_tools' argument
+def create_agent(llm, data):
+    """Create agent with tool names. create_pandas_dataframe_agent typically includes its own Python REPL tool."""
     return create_pandas_dataframe_agent(
         llm,
         data,
         agent_type="tool-calling",
         verbose=True,
         allow_dangerous_code=True,
-        extra_tools=tools,
+        # The PythonAstREPLTool was causing a duplicate declaration.
+        # create_pandas_dataframe_agent is expected to provide its own REPL for DataFrame interaction.
+        # extra_tools=tools, 
         return_intermediate_steps=True
     )
     
@@ -45,7 +49,7 @@ def _prompt_generator(question: str, chart_required: bool, csv_url: str):
                      2. **Data Integrity:** Ensure proper handling of null values to maintain accuracy and reliability.  
                      3. **Communication:** Provide concise, professional, and well-structured responses.
                      4. Avoid including any internal processing details or references to the methods used to generate your response (ex: based on the tool call, using the function -> These types of phrases.)
-                     5. Always use pd.read_csv({csv_url}) to read the CSV file.
+                     5. Always use pd.read_csv("{csv_url}") to read the CSV file.
 
                      **Query:** {question}  
                  """               
@@ -80,7 +84,7 @@ def _prompt_generator(question: str, chart_required: bool, csv_url: str):
                    - Use THE SAME unique_id throughout entire process
                    - NEVER generate new UUIDs after initial creation
                    - Return EXACT filepath string of the final saved chart
-                   - Always use pd.read_csv({csv_url}) to read the CSV file
+                   - Always use pd.read_csv("{csv_url}") to read the CSV file
                    """
 
     if chart_required:
@@ -98,21 +102,23 @@ def langchain_gemini_csv_handler(csv_url: str, question: str, chart_required: bo
             llm = llm_instances[current_instance_index]
             print(f"Using LLM instance index {current_instance_index}")
 
-            # Create tool with validated name
-            tool = PythonAstREPLTool(
-                locals={
-                    "df": data,
-                    "pd": pd,
-                    "np": np,
-                    "plt": plt,
-                    "sns": sns,
-                    "matplotlib": matplotlib,
-                    "uuid": uuid,
-                    "dt": dt
-                },
-            )
+            # Removed explicit creation of PythonAstREPLTool
+            # The create_pandas_dataframe_agent is expected to internally handle Python execution for DataFrame operations.
+            # tool = PythonAstREPLTool(
+            #     locals={
+            #         "df": data,
+            #         "pd": pd,
+            #         "np": np,
+            #         "plt": plt,
+            #         "sns": sns,
+            #         "matplotlib": matplotlib,
+            #         "uuid": uuid,
+            #         "dt": dt
+            #     },
+            # )
 
-            agent = create_agent(llm, data, [tool])
+            # Modified create_agent call: Removed the 'tools' argument
+            agent = create_agent(llm, data) 
             prompt = _prompt_generator(question, chart_required, csv_url)
             result = agent.invoke({"input": prompt})
             output = result.get("output")
@@ -127,5 +133,4 @@ def langchain_gemini_csv_handler(csv_url: str, question: str, chart_required: bo
             current_instance_index += 1
 
     print("All LLM instances have been exhausted.")
-    return None
-
+    return None