Added break down question via LLM.

README.md

@@ -1,5 +1,5 @@
 ---
-title: Template Final Assignment
+title: LLM-Enhanced Internet Search Agent
 emoji: 🕵🏻‍♂️
 colorFrom: indigo
 colorTo: indigo
@@ -12,4 +12,40 @@ hf_oauth: true
 hf_oauth_expiration_minutes: 480
 ---
 
+# LLM-Enhanced Internet Search Agent
+
+This agent uses a two-step approach to answer questions:
+
+1. **Question Breakdown**: The agent first uses an LLM (GPT-3.5) to break down complex questions into 2-3 key search queries
+2. **Targeted Search**: Each search query is sent to Wikipedia's API to retrieve relevant information
+3. **Answer Synthesis**: The agent then uses the LLM to synthesize a comprehensive answer based on all search results
+
+## Features
+
+- **Smart Query Generation**: Transforms natural language questions into optimized search queries
+- **Parallel Search Processing**: Searches for multiple key aspects of the question simultaneously
+- **Knowledge Synthesis**: Combines information from multiple sources into a cohesive answer
+- **Fallback Mechanisms**: Graceful handling of errors at each step of the process
+
+## Setup Requirements
+
+1. Clone this repository
+2. Install required packages: `pip install -r requirements.txt`
+3. Set your OpenAI API key as an environment variable: `OPENAI_API_KEY=your-api-key`
+
+## How It Works
+
+1. User submits a question
+2. LLM breaks down the question into key search terms
+3. Search terms are used to query Wikipedia API
+4. Results from multiple searches are collected
+5. LLM synthesizes the information into a comprehensive answer
+6. Answer is returned to the user
+
+This approach is more effective than direct internet searches because:
+- It identifies the most relevant aspects of complex questions
+- It can break multi-part questions into their components
+- It leverages the LLM's understanding of natural language
+- It provides more targeted and accurate search results
+
 Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference

app.py

@@ -3,6 +3,7 @@ import gradio as gr
 import requests
 import inspect
 import pandas as pd
+import openai  # Import OpenAI library
 
 # (Keep Constants as is)
 # --- Constants ---
@@ -13,6 +14,54 @@ DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space"
 class BasicAgent:
     def __init__(self):
         print("BasicAgent initialized.")
+        # Initialize OpenAI client - you'll need to set OPENAI_API_KEY in environment variables
+        self.openai_client = openai.OpenAI(api_key=os.getenv("OPENAI_API_KEY"))
+        if not os.getenv("OPENAI_API_KEY"):
+            print("Warning: OPENAI_API_KEY not found in environment variables.")
+    
+    def break_down_question(self, question: str) -> list:
+        """
+        Use an LLM to break down a complex question into key search terms or sub-questions.
+        
+        Args:
+            question (str): The original question
+            
+        Returns:
+            list: A list of key search terms or sub-questions
+        """
+        try:
+            print(f"Breaking down question with LLM: {question[:50]}...")
+            
+            # Create a prompt that asks the LLM to break down the question
+            prompt = f"""
+            Please break down this question into 2-3 key search queries that would help find information to answer it.
+            Return ONLY the search queries, one per line, with no additional text or explanations.
+            
+            Question: {question}
+            """
+            
+            # Call the OpenAI API to get the breakdown
+            response = self.openai_client.chat.completions.create(
+                model="gpt-3.5-turbo",
+                messages=[
+                    {"role": "system", "content": "You are a helpful assistant that breaks down questions into key search terms."},
+                    {"role": "user", "content": prompt}
+                ],
+                temperature=0.3,
+                max_tokens=150
+            )
+            
+            # Extract the search terms from the response
+            search_terms = response.choices[0].message.content.strip().split('\n')
+            search_terms = [term.strip() for term in search_terms if term.strip()]
+            
+            print(f"Question broken down into {len(search_terms)} search terms: {search_terms}")
+            return search_terms
+            
+        except Exception as e:
+            print(f"Error breaking down question: {e}")
+            # If there's an error, return the original question as a fallback
+            return [question]
         
     def search_internet(self, query: str) -> str:
         """
@@ -85,19 +134,59 @@ class BasicAgent:
     def __call__(self, question: str) -> str:
         print(f"Agent received question (first 50 chars): {question[:50]}...")
         
-        # Use internet search to find answer
-        search_results = self.search_internet(question)
+        # Use LLM to break down the question into key search terms
+        search_terms = self.break_down_question(question)
         
-        # Create a response based on search results
-        if (search_results and search_results != "No relevant information found." and 
-            not search_results.startswith("Error")):
-            answer = f"Based on my search, I found this information:\n\n{search_results}"
+        # Search for information using each search term
+        all_results = []
+        for term in search_terms:
+            result = self.search_internet(term)
+            if result and result != "No relevant information found." and not result.startswith("Error"):
+                all_results.append(result)
+        
+        # Create a response based on collected search results
+        if all_results:
+            # Join the results with clear separation
+            combined_results = "\n\n--- Next Search Result ---\n\n".join(all_results)
+            
+            # Use LLM to synthesize a coherent answer from the search results
+            try:
+                synthesis_prompt = f"""
+                Based on the following search results, please provide a comprehensive answer to this question:
+                
+                Question: {question}
+                
+                Search Results:
+                {combined_results}
+                
+                Answer:
+                """
+                
+                response = self.openai_client.chat.completions.create(
+                    model="gpt-3.5-turbo",
+                    messages=[
+                        {"role": "system", "content": "You are a helpful assistant that synthesizes information to answer questions accurately."},
+                        {"role": "user", "content": synthesis_prompt}
+                    ],
+                    temperature=0.5,
+                    max_tokens=500
+                )
+                
+                answer = response.choices[0].message.content.strip()
+                print("Agent returning synthesized answer from search results.")
+                return answer
+            
+            except Exception as e:
+                print(f"Error synthesizing answer: {e}")
+                # Fallback to returning the raw search results
+                answer = f"Based on my searches, I found this information:\n\n{combined_results}"
+                print("Agent returning raw search results due to synthesis error.")
+                return answer
         else:
-            # Fallback to default answer if search fails
+            # Fallback to default answer if all searches fail
             answer = "I couldn't find specific information about that question."
-        
-        print(f"Agent returning answer based on internet search.")
-        return answer
+            print("Agent returning default answer as searches found no useful information.")
+            return answer
 
 def run_and_submit_all( profile: gr.OAuthProfile | None):
     """
@@ -222,7 +311,7 @@ def run_and_submit_all( profile: gr.OAuthProfile | None):
 
 # --- Build Gradio Interface using Blocks ---
 with gr.Blocks() as demo:
-    gr.Markdown("# Basic Agent Evaluation Runner (Attempt #1)")
+    gr.Markdown("# Basic Agent Evaluation Runner (Attempt #2)")
     gr.Markdown(
         """
         **Instructions:**

requirements.txt

@@ -1,2 +1,3 @@
 gradio
-requests
+requests
+openai