Process flow visualizer + key skills [for validation only) V6

integration_validation.py

@@ -0,0 +1 @@
+ 

src/agents/skills_identification_agent.py

@@ -130,11 +130,18 @@ class SkillsIdentificationAgent:
             skill_classification = await self._classify_skills(user_input, context)
             
             # Combine results
+            combined_data = {
+                "market_analysis": market_analysis,
+                "skill_classification": skill_classification,
+                "user_input": user_input,
+                "context": context
+            }
+            
             result = {
                 "agent_id": self.agent_id,
                 "market_analysis": market_analysis,
                 "skill_classification": skill_classification,
-                "identified_skills": self._extract_high_probability_skills(skill_classification),
+                "identified_skills": self._extract_high_probability_skills(combined_data),
                 "processing_time": market_analysis.get("processing_time", 0) + skill_classification.get("processing_time", 0),
                 "confidence_score": self._calculate_overall_confidence(market_analysis, skill_classification)
             }
@@ -423,7 +430,8 @@ Respond in JSON format:
         high_prob_skills = []
         
         # From market analysis
-        market_skills = classification.get("market_analysis", {}).get("selected_skills", [])
+        market_analysis = classification.get("market_analysis", {})
+        market_skills = market_analysis.get("selected_skills", [])
         for skill in market_skills:
             if skill.get("relevance_score", 0) > 0.2:
                 high_prob_skills.append({
@@ -434,7 +442,8 @@ Respond in JSON format:
                 })
         
         # From skill classification
-        classification_skills = classification.get("skill_classification", {}).get("top_skills", [])
+        skill_classification = classification.get("skill_classification", {})
+        classification_skills = skill_classification.get("top_skills", [])
         for skill in classification_skills:
             if skill.get("probability", 0) > 0.2:
                 high_prob_skills.append({
@@ -444,6 +453,30 @@ Respond in JSON format:
                     "source": "skill_classification"
                 })
         
+        # If no skills found from LLM, use rule-based fallback
+        if not high_prob_skills:
+            logger.warning(f"{self.agent_id} No skills identified from LLM, using rule-based fallback")
+            # Extract user input from context if available
+            user_input = ""
+            if isinstance(classification, dict) and "user_input" in classification:
+                user_input = classification["user_input"]
+            elif isinstance(classification, dict) and "context" in classification:
+                context = classification["context"]
+                if isinstance(context, dict) and "user_input" in context:
+                    user_input = context["user_input"]
+            
+            if user_input:
+                rule_based_result = self._rule_based_skill_classification(user_input)
+                rule_skills = rule_based_result.get("top_skills", [])
+                for skill in rule_skills:
+                    if skill.get("probability", 0) > 0.2:
+                        high_prob_skills.append({
+                            "skill": skill["skill"],
+                            "category": "rule_based",
+                            "probability": skill["probability"],
+                            "source": "rule_based_fallback"
+                        })
+        
         # Remove duplicates and sort by probability
         unique_skills = {}
         for skill in high_prob_skills:

src/llm_router.py

@@ -78,8 +78,9 @@ class LLMRouter:
             import requests
             
             model_id = model_config["model_id"]
+            is_chat_model = model_config.get("is_chat_model", True)
             
-            # Use the chat completions endpoint
+            # Use the chat completions endpoint for chat models
             api_url = "https://router.huggingface.co/v1/chat/completions"
             
             logger.info(f"Calling HF Chat Completions API for model: {model_id}")

src/models_config.py

@@ -8,27 +8,31 @@ LLM_CONFIG = {
             "max_tokens": 2000,
             "temperature": 0.7,
             "cost_per_token": 0.000015,
-            "fallback": "gpt2"  # Simple but guaranteed working model
+            "fallback": "gpt2",  # Simple but guaranteed working model
+            "is_chat_model": True
         },
         "embedding_specialist": {
             "model_id": "sentence-transformers/all-MiniLM-L6-v2",
             "task": "embeddings",
             "vector_dimensions": 384,
             "purpose": "semantic_similarity",
-            "cost_advantage": "90%_cheaper_than_primary"
+            "cost_advantage": "90%_cheaper_than_primary",
+            "is_chat_model": False
         },
         "classification_specialist": {
-            "model_id": "cardiffnlp/twitter-roberta-base-emotion",
+            "model_id": "Qwen/Qwen2.5-7B-Instruct",  # Use chat model for classification
             "task": "intent_classification",
             "max_length": 512,
             "specialization": "fast_inference",
-            "latency_target": "<100ms"
+            "latency_target": "<100ms",
+            "is_chat_model": True
         },
         "safety_checker": {
-            "model_id": "unitary/unbiased-toxic-roberta",
+            "model_id": "Qwen/Qwen2.5-7B-Instruct",  # Use chat model for safety
             "task": "content_moderation",
             "confidence_threshold": 0.85,
-            "purpose": "bias_detection"
+            "purpose": "bias_detection",
+            "is_chat_model": True
         }
     },
     "routing_logic": {