Perform some cleanup

.env.example

@@ -10,5 +10,6 @@ LANGFUSE_HOST=<YOUR_LANGFUSE_HOST>
 TAVILY_API_KEY=<YOUR_TAVILY_KEY>
 
 # Run configurations
-USE_DDGS=False
-USE_RATE_LIMITER=True
+USE_DDGS=True
+USE_RATE_LIMITER=True
+RECURSION_LIMIT=<YOUR_RECURSION_LIMIT>

.gitignore

@@ -1,6 +1,5 @@
 .venv/
 .env
-# Python cache files
 __pycache__/
 .dist/
 test_*

README.md

@@ -1,15 +1,47 @@
+# 🕵️‍♂️ Super Agent: The AI Answer Machine 🚀
+
+Welcome to **Super Agent** – the AI agent built to be evaluated on [GAAI](https://huggingface.co/papers/2311.12983) benchmark to complete the [AI agents](https://huggingface.co/learn/agents-course/unit0/introduction) course by [Hugging Face](https://huggingface.co/)!
+
 ---
-title: Template Final Assignment
-emoji: 🕵🏻‍♂️
-colorFrom: indigo
-colorTo: indigo
-sdk: gradio
-sdk_version: 5.25.2
-app_file: app.py
-pinned: false
-hf_oauth: true
-# optional, default duration is 8 hours/480 minutes. Max duration is 30 days/43200 minutes.
-hf_oauth_expiration_minutes: 480
+
+## ✨ Features
+
+- 🔍 **Web Search**: Finds up-to-date info using DuckDuckGo or Tavily.
+- 📊 **Excel & Data Extraction**: Reads and analyzes Excel files.
+- 🎧 **Audio & YouTube Transcripts**: Extracts transcripts from audio files and YouTube videos.
+- 🧮 **Math Tools**: Adds numbers, checks commutativity, and more.
+- 🧠 **RAG-Ready**: Retrieval-Augmented Generation for smarter, more accurate answers.
+- 🤖 **LLM Integration**: Works with Google Gemini, HuggingFace, and OpenAI models.
+- 🛡️ **Exact Match Mode**: Answers are evaluated for exactness.
+- 🛠️ **Extensible Tooling**: Easily add new tools for more superpowers.
+- ⏳ **Gemini Rate Limiter**: Includes a rate limiter implementation for Gemini, perfect for handling free tier API limits.
+
+---
+
+## 🚀 Quickstart
+
+1. **Clone the repo**  
+   `git clone https://github.com/yourusername/super_agent.git`
+
+2. **Install dependencies**  
+   `pip install -r requirements.txt`
+
+3. **Set up your `.env` file**  
+   Copy `.env.example` to `.env` and fill in your API keys.
+
+4. **Run the app**  
+   `python app.py`
+
+5. **Test the agent**  
+   Try the test scripts or use the Gradio interface!
+
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+## 🧩 Supported Tools
+
+- `web_search` – Search the web for up-to-date info
+- `extract_data_from_excel` – Download and parse Excel files
+- `extract_transcript_from_youtube` – Get YouTube video transcripts
+- `extract_transcript_from_audio` – Get transcripts from audio files
+- `add_numbers_in_list` – Add up numbers
+- `check_commutativity` – Test if a binary operation is commutative

agent.py

@@ -1,7 +1,5 @@
 import os
-import getpass
 import requests
-import datetime
 from dotenv import load_dotenv
 from langgraph.graph import StateGraph, MessagesState, START
 from langchain_huggingface import ChatHuggingFace, HuggingFaceEndpoint
@@ -150,14 +148,14 @@ if __name__ == "__main__":
     # task_id = "cabe07ed-9eca-40ea-8ead-410ef5e83f91" # Louvrier ?
     # task_id = "f918266a-b3e0-4914-865d-4faa564f1aef"  # Code example
     # task_id = "3f57289b-8c60-48be-bd80-01f8099ca449" # at bats ?
-    task_id = "7bd855d8-463d-4ed5-93ca-5fe35145f733" # Excel file (passed)
-    # task_id = "5a0c1adf-205e-4841-a666-7c3ef95def9d" # Malko competition (PASS)
-    # task_id = "305ac316-eef6-4446-960a-92d80d542f82" # Poland film (FAIL)
-    # task_id = "bda648d7-d618-4883-88f4-3466eabd860e"  # Vietnamese (FAIL)
+    task_id = "7bd855d8-463d-4ed5-93ca-5fe35145f733" # Excel file
+    # task_id = "5a0c1adf-205e-4841-a666-7c3ef95def9d" # Malko competition
+    # task_id = "305ac316-eef6-4446-960a-92d80d542f82" # Poland film
+    # task_id = "bda648d7-d618-4883-88f4-3466eabd860e"  # Vietnamese
     # task_id = "cf106601-ab4f-4af9-b045-5295fe67b37d" # Olympics
     # task_id = "a0c07678-e491-4bbc-8f0b-07405144218f" # pitchers
     # task_id = "3cef3a44-215e-4aed-8e3b-b1e3f08063b7" # grocery list
-    # task_id = "840bfca7-4f7b-481a-8794-c560c340185d" # Carolyn Collins Petersen (FAIL)
+    # task_id = "840bfca7-4f7b-481a-8794-c560c340185d" # Carolyn Collins Petersen
     # task_id = "1f975693-876d-457b-a649-393859e79bf3" # Audio (pages)
     # task_id = "99c9cc74-fdc8-46c6-8f8d-3ce2d3bfeea3" # Audio (recipe)
 

app.py

@@ -1,8 +1,6 @@
 """
 NOTE:
-    - The agent only runs on a subset of tasks to avoid unnecessary token/api usage for questions that the agent
-      cannot handle right now. The task ids to exclude are in the `excluded_tasks.txt` file.
-    - There is a 30 sec delay after each question is answered to avoid rate limiting issues.
+    - If USE_RATE_LIMITER env variable is True, the agent will use a rate limiter to avoid hitting API limits.
 """
 
 import os
@@ -11,7 +9,6 @@ import requests
 import inspect
 import pandas as pd
 import datetime
-import time
 from agent import build_agent
 from langchain_core.messages import HumanMessage
 from langfuse.langchain import CallbackHandler
@@ -30,6 +27,7 @@ class SuperAgent:
     def __init__(self):
         print("SuperAgent initialized.")
         self.agent = build_agent(provider="google")  # Change to "hf" for HuggingFace
+        self.recursion_limit = os.getenv("RECURSION_LIMIT", "25")
 
     def __call__(self, data: dict) -> str:
         """
@@ -86,7 +84,7 @@ class SuperAgent:
         try:
             answer = self.agent.invoke(
                 {"messages": [human_msg]},
-                config={"callbacks": [langfuse_handler], "recursion_limit": 15}
+                config={"callbacks": [langfuse_handler], "recursion_limit": self.recursion_limit}
             )
 
             # for message in answer["messages"]:
@@ -184,11 +182,6 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
         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}"})
-        
-        # If it's not the last question sleep
-        # if idx < len(questions_data) - 1:
-        #     print(f"[{datetime.datetime.now().strftime("%Y-%m-%d %H:%M:%S")}] Sleeping for 60 seconds to avoid rate limiting issues...")
-        #     time.sleep(60) # Sleep to avoid rate limiting issues
 
     if not answers_payload:
         print("Agent did not produce any answers to submit.")
@@ -248,11 +241,9 @@ with gr.Blocks() as demo:
     gr.Markdown("# Super Agent Evaluation Runner")
     gr.Markdown(
         """
-        Welcome to my **Super Agent** Runner!
+        ## Welcome to my **Super Agent** Runner!
         
-        Please, sit back and relax while the evaluation runs!
-
-        There is a 60 second delay after each question is answered to avoid rate limiting issues.
+        Please, sit back and relax while the evaluation runs on the GAAI benchmark!
         """
     )
 

rate_limiters.py

@@ -3,6 +3,26 @@ import datetime
 from google.api_core.exceptions import ResourceExhausted
 
 def safe_invoke_with_retry_gemini(llm_with_tools, messages, max_retries=3, wait_seconds=60):
+    """
+    Safely invokes a Gemini LLM with automatic retries on rate limit errors.
+
+    This function attempts to call the provided LLM with the given messages. If a ResourceExhausted
+    (rate limit) error occurs, it waits for a specified number of seconds and retries, up to a maximum
+    number of retries. Other exceptions are raised immediately.
+
+    Args:
+        llm_with_tools: The Gemini LLM instance with tools bound.
+        messages (list): List of messages to send to the LLM.
+        max_retries (int): Maximum number of retry attempts on rate limit errors.
+        wait_seconds (int): Seconds to wait between retries.
+
+    Returns:
+        The result of llm_with_tools.invoke(messages) if successful.
+
+    Raises:
+        ResourceExhausted: If the maximum number of retries is reached due to rate limiting.
+        Exception: Any other exception encountered during invocation.
+    """
     for attempt in range(1, max_retries + 1):
         try:
             return llm_with_tools.invoke(messages)

tools/web_search.py

@@ -138,7 +138,7 @@ def web_search(query: str) -> str:
         chunks (str): Concatenated string of most relevant chunks.
     """
 
-    USE_DDGS = os.getenv("USE_DDGS").lower() == "true"
+    USE_DDGS = os.getenv("USE_DDGS", "false").lower() == "true"
     # ----- STEP 1: Find the most relevant webpages
     if USE_DDGS:
         results = DDGS(timeout=30).text(query, max_results=MAX_RESULTS)