Merge branch 'feat/agent-improvement'

ShrewdAgent.py

@@ -1,4 +1,5 @@
 import os
+from textwrap import dedent
 from typing import TypedDict, Annotated, Optional, Any, Callable, Sequence, Union
 
 from langchain_core.messages import AnyMessage, SystemMessage, HumanMessage
@@ -29,18 +30,21 @@ class AgentState(TypedDict):
 
 
 class ShrewdAgent:
-    message_system = """You are a general AI assistant equipped with a suite of external tools. Your task is to
-                        answer the following question as accurately and helpfully as possible by using the tools
-                        provided. Do not write or execute code yourself. For any operation requiring computation,
-                        data retrieval, or external access, explicitly invoke the appropriate tool.
+    message_system = dedent("""
+        You are a general AI assistant equipped with a suite of external tools. Your task is to
+        answer the following question as accurately and helpfully as possible by using the tools
+        provided. Do not write or execute code yourself. For any operation requiring computation,
+        data retrieval, or external access, explicitly invoke the appropriate tool.
         
-                        Follow these guidelines:"
-                        - Clearly explain your reasoning step by step.
-                        - Justify your choice of tool(s) at each step.
-                        - If multiple interpretations are possible, outline them and explain your reasoning for selecting one.
-                        - If the answer requires external data or inference, retrieve or deduce it via the available tools.
-        
-                        Important: Your final output must be only a number or a short phrase, with no additional text or explanation."""
+        Follow these guidelines:
+        - Clearly explain your reasoning step by step.
+        - Justify your choice of tool(s) at each step.
+        - If multiple interpretations are possible, outline them and explain your reasoning for selecting one.
+        - If the answer requires external data or inference, retrieve or deduce it via the available tools.
+
+        Important: Your final output MUST be only a number or a word with no additional text or explanation, 
+        unless the response format is explicitly specified in the question. Do not include reasoning, 
+        commentary, or any other content beyond the requested answer.""")
 
     def __init__(self):
         self.tools = [
@@ -56,7 +60,7 @@ class ShrewdAgent:
             transcribe_audio,
         ]
         self.llm = ChatOpenAI(
-            model="gpt-4o-mini",
+            model="gpt-4.1",
             temperature=0,
             api_key=SecretStr(os.environ['OPENAI_API_KEY'])
         ).bind_tools(self.tools)
@@ -67,7 +71,8 @@ class ShrewdAgent:
             }
 
         self.agent = _build_state_graph(AgentState, assistant_node, self.tools)
-        logger.info("ShrewdAgent initialized.")
+        logger.info(f"Agent initialized with tools: {[tool.name for tool in self.tools]}")
+        logger.debug(f"system message:\n{self.message_system}")
 
     def __call__(self, question: str) -> str:
         logger.info(f"Agent received question:\n{question}")

app.py

@@ -1,4 +1,5 @@
 import os
+import time
 
 import gradio as gr
 import pandas as pd
@@ -87,6 +88,7 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
             submitted_answer = agent(question_with_attachment)
             answers_payload.append({"task_id": task_id, "submitted_answer": submitted_answer})
             results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": submitted_answer})
+            time.sleep(70) # wait for reducing rate limit errors
         except Exception as e:
              print(f"Error running agent on task {task_id}: {e}")
              results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": f"AGENT ERROR: {e}"})