Final 7.2.3

README.md

@@ -1,8 +1,8 @@
 ---
 title: GAIA Agent System
 emoji: 🤖
-colorFrom: indigo
-colorTo: indigo
+colorFrom: yellow
+colorTo: pink
 sdk: gradio
 sdk_version: 5.25.2
 app_file: ./src/app.py

src/agents/web_researcher.py

@@ -445,21 +445,35 @@ class WebResearchAgent:
             if len(phrase.strip()) > 0:
                 priority_terms.append(phrase.strip())
         
-        # Extract proper nouns (capitalized words)
+        # Extract years (4-digit numbers) - capture full years, not just prefixes
+        years = re.findall(r'\b(?:19|20)\d{2}\b', question)  # Changed from capturing group to full match
+        
+        # Extract proper nouns (capitalized words) - exclude numbers
         proper_nouns = []
         for word in question.split():
             clean_word = re.sub(r'[^\w]', '', word)
-            if clean_word and clean_word[0].isupper() and len(clean_word) > 1:
+            if (clean_word and 
+                clean_word[0].isupper() and 
+                len(clean_word) > 1 and 
+                not clean_word.isdigit()):  # Exclude pure numbers
                 proper_nouns.append(clean_word)
         
-        # Extract years (4-digit numbers)
-        years = re.findall(r'\b(19|20)\d{2}\b', question)
-        
-        # Extract other important numbers (but not random ones)
-        important_numbers = re.findall(r'\b\d{1,4}\b', question)
-        # Filter out years and common numbers from important numbers to avoid duplication
-        common_numbers = {'19', '20', '1', '2', '3', '4', '5', '10'}  # Filter out very common numbers
-        important_numbers = [num for num in important_numbers if num not in years and num not in common_numbers]
+        # Extract other meaningful numbers (but be very selective)
+        # Only include numbers that are likely meaningful (dates, counts, etc.)
+        meaningful_numbers = []
+        number_matches = re.findall(r'\b\d{1,4}\b', question)
+        for num in number_matches:
+            # Skip very common/meaningless numbers and years already captured
+            if (num not in ['1', '2', '3', '4', '5', '10', '20', '19', '21', '22', '23', '24', '25'] and
+                num not in years and 
+                len(num) > 1):  # Require at least 2 digits for meaningful numbers
+                # Only include if it appears in a meaningful context
+                if any(context in question.lower() for context in [
+                    f'{num} albums', f'{num} songs', f'{num} years', f'{num} people', 
+                    f'{num} times', f'{num} days', f'{num} months', f'episode {num}',
+                    f'season {num}', f'volume {num}', f'part {num}'
+                ]):
+                    meaningful_numbers.append(num)
         
         # Build search terms with priority
         search_terms = []
@@ -492,7 +506,7 @@ class WebResearchAgent:
         
         # Add a few important numbers if space allows
         if len(' '.join(search_terms)) < max_length - 10:
-            search_terms.extend(important_numbers[:2])
+            search_terms.extend(meaningful_numbers[:2])
         
         # Join and clean up
         search_query = ' '.join(search_terms)

src/production_deployment_guide.md

@@ -312,3 +312,4 @@ With proper deployment and authentication:
 - **Deployment**: Ready for immediate HuggingFace Space deployment
 
 **The GAIA Agent is now a focused, high-performance system using proper AI models and multi-agent orchestration!** 🎉
+ 

src/tools/final_answer_tool.py

@@ -55,7 +55,7 @@ class FinalAnswerTool:
                 }
             
             # Parse and clean the extracted answer
-            extracted_answer = self._clean_answer(result.response, question_type)
+            extracted_answer = self._clean_answer(result.response, question, question_type)
             
             # Validate answer format
             validation_result = self._validate_answer(extracted_answer, question_type)
@@ -141,7 +141,7 @@ Extract the precise answer NOW:"""
         
         return base_prompt
     
-    def _clean_answer(self, raw_answer: str, question_type: str) -> str:
+    def _clean_answer(self, raw_answer: str, question: str, question_type: str) -> str:
         """Clean and format the extracted answer"""
         
         # Remove common unwanted prefixes/suffixes

src/tools/web_search_tool.py

@@ -229,13 +229,27 @@ class WebSearchTool(BaseTool):
     
     def _search_with_duckduckgo(self, query: str, limit: int = 5, extract_content: bool = False) -> Dict[str, Any]:
         """
-        Search using DuckDuckGo - primary search engine with improved error handling
+        Search using DuckDuckGo - primary search engine with improved error handling and rate limiting
         """
         try:
             logger.info(f"🦆 DuckDuckGo search for: {query}")
             
-            # Use DuckDuckGo text search - fail fast if there are issues
-            ddg_results = list(self.ddgs.text(query, max_results=min(limit, 10)))
+            # Add small delay to avoid rate limiting
+            time.sleep(0.5)
+            
+            # Use DuckDuckGo text search with retry logic
+            max_retries = 2
+            for attempt in range(max_retries):
+                try:
+                    ddg_results = list(self.ddgs.text(query, max_results=min(limit, 10)))
+                    break
+                except Exception as retry_error:
+                    if attempt < max_retries - 1:
+                        logger.warning(f"DuckDuckGo attempt {attempt + 1} failed, retrying in {2 ** attempt}s: {retry_error}")
+                        time.sleep(2 ** attempt)  # Exponential backoff
+                        continue
+                    else:
+                        raise retry_error
             
             if not ddg_results:
                 logger.warning("DuckDuckGo returned no results")
@@ -264,7 +278,10 @@ class WebSearchTool(BaseTool):
             
         except Exception as e:
             logger.warning(f"DuckDuckGo search failed: {str(e)}")
-            # Don't log the full exception details to avoid spam
+            # Check if it's a rate limiting error and add longer delay
+            if "ratelimit" in str(e).lower() or "429" in str(e) or "202" in str(e):
+                logger.warning("Rate limiting detected, adding delay before fallback")
+                time.sleep(2.0)
             return self._search_with_fallback(query, limit)
     
     def _search_with_fallback(self, query: str, limit: int = 5) -> Dict[str, Any]:
@@ -304,7 +321,7 @@ class WebSearchTool(BaseTool):
         # Fall back to Wikipedia search
         logger.info("📚 Wikipedia search for: " + query)
         try:
-            wiki_results = self._search_wikipedia(query, limit)
+            wiki_results = self._search_with_wikipedia(query, limit)
             if wiki_results and wiki_results.get('success'):
                 logger.info(f"✅ Wikipedia found {wiki_results.get('count', 0)} results")
                 return wiki_results