feat: introduce MultiStepWorkflow for enhanced agent functionality and update agent initialization

agent.py

@@ -1,4 +1,5 @@
 import asyncio
+import datetime
 import os
 import pprint
 from llama_index.core.agent.workflow.workflow_events import AgentInput, AgentOutput, AgentStream, ToolCall, ToolCallResult
@@ -15,9 +16,6 @@ from llama_index.llms.nebius import NebiusLLM
 import tools
 
 """
-TODO: 3. Fix submission error 'Agent finished. Submitting 13 answers for user 'agsagds'...
-Submitting 13 answers to: https://agents-course-unit4-scoring.hf.space/submit
-An unexpected error occurred during submission: Object of type AgentOutput is not JSON serializable'
 TODO: 4. Run HF Space and submit answers to the scoring server.
 TODO: 5. Get a certificate for the course.
 """
@@ -25,20 +23,6 @@ TODO: 5. Get a certificate for the course.
 class BasicAgent:
     def __init__(self, verbose: bool = False):
         self.verbose = verbose
-        self.system_prompt = """
-        You are a general AI assistant. I will ask you a question. 
-        Report your thoughts, and finish your answer with the following template: 
-        FINAL ANSWER: [YOUR FINAL ANSWER]. 
-        YOUR FINAL ANSWER should be a number OR as few words as possible 
-        OR a comma separated list of numbers and/or strings. 
-        If you are asked for a number, don't use comma to write your number 
-        neither use units such as $ or percent sign unless specified otherwise. 
-        If you are asked for a string, don't use articles, neither abbreviations 
-        (e.g. for cities), and write the digits in plain text unless specified 
-        otherwise. If you are asked for a comma separated list, apply the above 
-        rules depending of whether the element to be put in the list is a number 
-        or a string.
-        """
         # llm = Ollama(
         #     model="gpt-oss:20b",
         #     request_timeout=120.0
@@ -63,13 +47,12 @@ class BasicAgent:
                 *WikipediaToolSpec().to_tool_list(),
                 ],
             llm=llm,
-            system_prompt=self.system_prompt
+            system_prompt=f"Today's date is {datetime.datetime.now().strftime('%Y-%m-%d')}." 
         )
-        print("BasicAgent initialized.")
     
     async def __call__(self, question: str) -> str:
         """Async call method that returns the final result without streaming"""
-        print(f"Agent received question (first 50 chars): {question[:50]}...")
+        print(f"Agent received question : {question}\n")
         
         if self.verbose:
             answer = await self.stream_answers(question)
@@ -100,7 +83,7 @@ class BasicAgent:
             elif isinstance(event, ToolCall):
                 print(f"\n\tCalled tool: {event.tool_name} {event.tool_kwargs}")
             elif isinstance(event, ToolCallResult):
-                print(f"\n\t{event.tool_name} {event.tool_kwargs} -> {event.tool_output}")
+                print(f"\n\t{event.tool_name} {event.tool_kwargs} -> {str(event.tool_output)[200:]}")
             elif isinstance(event, StopEvent):
                 return event.result
             elif isinstance(event, AgentStream):

app.py

@@ -4,6 +4,7 @@ import requests
 import inspect
 import pandas as pd
 from agent import BasicAgent
+from workflow import MultiStepWorkflow
 
 # (Keep Constants as is)
 # --- Constants ---
@@ -97,10 +98,10 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
 
     # 1. Instantiate Agent ( modify this part to create your agent)
     try:
-        agent = BasicAgent(verbose=True)
+        agent_workflow = MultiStepWorkflow(timeout=300, verbose=False)
     except Exception as e:
-        print(f"Error instantiating agent: {e}")
-        return f"Error initializing agent: {e}", None
+        print(f"Error instantiating agent workflow: {e}")
+        return f"Error initializing agent workflow: {e}", None
     # In the case of an app running as a hugging Face space, this link points toward your codebase ( usefull for others so please keep it public)
     agent_code = f"https://huggingface.co/spaces/{space_id}/tree/main"
     print(agent_code)
@@ -129,7 +130,7 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
     # 3. Run your Agent
     results_log = []
     answers_payload = []
-    print(f"Running agent on {len(questions_data)} questions...")
+    print(f"Running agent workflow on {len(questions_data)} questions...")
     for item in questions_data:
         task_id = item.get("task_id")
         question_text = item.get("question")
@@ -146,22 +147,22 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
             print(f"Skipping item with missing task_id or question: {item}")
             continue
         try:
-            print(f"Running agent on question: {task_id} {question_text}")
-            submitted_answer = agent.call_sync(question_text)
+            print(f"Running agent workflow on question: {task_id} {question_text}")
+            submitted_answer = agent_workflow.callSync(question_text)
             print(f"Submitted answer: {submitted_answer}")
             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})
         except Exception as e:
-            print(f"Error running agent on task {task_id}: {e}")
+            print(f"Error running agent workflow on task {task_id}: {e}")
             results_log.append({"Task ID": task_id, "Question": question_text, "Submitted Answer": f"AGENT ERROR: {e}"})
 
     if not answers_payload:
-        print("Agent did not produce any answers to submit.")
-        return "Agent did not produce any answers to submit.", pd.DataFrame(results_log)
+        print("Agent workflow did not produce any answers to submit.")
+        return "Agent workflow did not produce any answers to submit.", pd.DataFrame(results_log)
 
     # 4. Prepare Submission 
     submission_data = {"username": username.strip(), "agent_code": agent_code, "answers": answers_payload}
-    status_update = f"Agent finished. Submitting {len(answers_payload)} answers for user '{username}'..."
+    status_update = f"Agent workflow finished. Submitting {len(answers_payload)} answers for user '{username}'..."
     print(status_update)
 
     # 5. Submit

workflow.py

@@ -0,0 +1,133 @@
+import asyncio
+import os
+import datetime
+from llama_index.core.workflow import Event, StartEvent, StopEvent, Workflow, step, Context
+from llama_index.llms.nebius import NebiusLLM
+
+from agent import BasicAgent
+
+"""
+1. Decide what to use: context or event payload?
+2. Investigate alternative to initializate llm at each step|global llm
+3. Throw relevant data throght steps
+"""
+class EvalPlanEvent(Event):
+    planStep: str
+
+class FinalAnswerEvent(Event):
+    finalAnswer: str
+
+class UpdatePlanEvent(Event):
+    planStep: str
+    planStepResult: str
+
+class MultiStepWorkflow(Workflow):
+    
+    def __init__(self, **kwargs):
+        self.llm = NebiusLLM(
+                api_key=os.getenv("NEBIUS_API_KEY"),
+                model="Qwen/Qwen3-235B-A22B-Instruct-2507",
+                api_base="https://api.tokenfactory.nebius.com/v1",
+                system_prompt=f"Today's date is {datetime.datetime.now().strftime('%Y-%m-%d')}." 
+            )
+        self.agent = BasicAgent(verbose=kwargs.get("verbose", False))
+        super().__init__(**kwargs)
+    
+    @step
+    async def makePlanStep(self, ctx: Context, ev: StartEvent) -> EvalPlanEvent:
+        if not hasattr(ev, "question"):
+            raise ValueError("question field is required")
+        await ctx.store.set("question", ev.question)
+        plan = await self.llm.acomplete("""Make a plan to answer the question. Plan should be a list of steps. 
+                                        Each step should contain enough context to execute the step.
+                                        Formulate the steps in a way that can be executed by the agent.
+                                        Maximum 7 steps. Return only the plan, no other text.
+                                        The question: """ + ev.question)
+        await ctx.store.set("plan", str(plan))
+        step = await self.llm.acomplete("""Get the first step of the plan. Return only the step, no other text.
+                                        The plan: """ + str(plan))
+        print(f'Plan is {plan}')
+        return EvalPlanEvent(planStep=str(step))
+
+    @step
+    async def evalPlanStep(self, ctx: Context, ev: EvalPlanEvent) -> UpdatePlanEvent | FinalAnswerEvent:
+        if not hasattr(ev, "planStep"):
+            raise ValueError("planStep field is required")
+        
+        result = await self.agent(f"The question is: {await ctx.store.get('question')} \n\n The plan is: {await ctx.store.get('plan')} \n\n Execute only the step: {ev.planStep}")
+        return UpdatePlanEvent(planStep=ev.planStep, planStepResult=str(result))
+
+    @step
+    async def updatePlanStep(self, ctx: Context, ev: UpdatePlanEvent)-> EvalPlanEvent | FinalAnswerEvent:
+        if not hasattr(ev, "planStep"):
+            raise ValueError("planStep field is required")
+        if not hasattr(ev, "planStepResult"):
+            raise ValueError("planStepResult field is required")
+        
+        plan = await ctx.store.get("plan")
+        question = await ctx.store.get("question")
+        plan = await self.llm.acomplete("""Update the plan based on the plan step result. 
+                                        Note that each plan step should contain enough context to execute the step and formulate the steps in a way that can be executed by the agent.
+                                        Maximum 7 steps.
+                                        Return only the updated plan, no other text.
+                                        The plan step: """ + ev.planStep + """
+                                        The plan step result: """ + ev.planStepResult + """
+                                        The question: """ + question + """
+                                        The plan: """ + plan)
+        await ctx.store.set("plan", str(plan))
+        
+        verdict = await self.llm.acomplete("""Check the question and the plan. If there is enough information to answer the question, then return answer with the following template: FINAL ANSWER: [FINAL ANSWER]. 
+                                        Otherwise return an empty string.
+                                        The question: """ + question + """
+                                        The plan: """ + str(plan))
+        print(f'Plan is {plan} \n\nVerdict is {verdict}')
+        if "FINAL ANSWER" in str(verdict):
+            return FinalAnswerEvent(finalAnswer=str(verdict).split("FINAL ANSWER:")[1])
+        
+        step = await self.llm.acomplete("""Get the next step to evaluate of the plan. Return only the step, no other text.
+                                        The plan: """ + str(plan))
+        return EvalPlanEvent(planStep=str(step))
+        
+    
+    @step
+    async def finalAnswerStep(self, ctx: Context, ev: FinalAnswerEvent) -> StopEvent:
+        if not hasattr(ev, "finalAnswer"):
+            raise ValueError("finalAnswer field is required")
+        
+        question = await ctx.store.get("question")
+        formattedAnswer = await self.llm.acomplete("""
+        Help me to format the answer to the correct format. Return only the formatted answer, no other text.
+        <question>
+        """ + question + """
+        </question>
+        <answer>
+        """ + ev.finalAnswer + """
+        </answer>
+        <formatting rules>
+        Answer should be a number OR as few words as possible OR a comma separated list of numbers and/or strings. 
+        If you are asked for a number, don't use comma to write your number neither use units such as $ or percent sign unless specified otherwise. 
+        If you are asked for a string, don't use articles, neither abbreviations (e.g. for cities), and write the digits in plain text unless specified otherwise. 
+        If you are asked for a comma separated list, apply the above rules depending of whether the element to be put in the list is a number  or a string. 
+        </formatting rules>"""
+        )
+        
+        return StopEvent(result=str(formattedAnswer))
+
+    def callSync(self, question: str) -> str:
+        return asyncio.run(self.run(question=question))
+
+if __name__ == "__main__":
+    async def main():
+        questionAlbums = "How many studio albums were published by Mercedes Sosa between 2000 and 2009 (included)? You can use the latest 2022 version of english wikipedia."
+        questionReverse = ".rewsna eht sa \"tfel\" drow eht fo etisoppo eht etirw ,ecnetnes siht dnatsrednu uoy fI"
+        questionDinosaur = "Who nominated the only Featured Article on English Wikipedia about a dinosaur that was promoted in November 2016?"
+        questionTable = "Given this table defining * on the set S = {a, b, c, d, e}\n\n|*|a|b|c|d|e|\n|---|---|---|---|---|---|\n|a|a|b|c|b|d|\n|b|b|c|a|e|c|\n|c|c|a|b|b|a|\n|d|b|e|b|e|d|\n|e|d|b|a|d|c|\n\nprovide the subset of S involved in any possible counter-examples that prove * is not commutative. Provide your answer as a comma separated list of the elements in the set in alphabetical order."
+        questionSurname = "What is the surname of the equine veterinarian mentioned in 1.E Exercises from the chemistry materials licensed by Marisa Alviar-Agnew & Henry Agnew under the CK-12 license in LibreText's Introductory Chemistry materials as compiled 08/21/2023?"
+        
+        from dotenv import load_dotenv
+        load_dotenv()
+        workflow = MultiStepWorkflow(timeout=300, verbose=False)
+        response = await workflow.run(question=questionSurname)
+        print(response)
+
+    asyncio.run(main())