Create utils/rag_agent.py

utils/rag_agent.py

@@ -0,0 +1,99 @@
+from sentence_transformers import SentenceTransformer
+from faiss import IndexFlatIP
+import numpy as np
+from huggingface_hub import InferenceClient
+import json
+import os
+
+class RAGInterviewAgent:
+    def __init__(self, job_role, cv_summary):
+        self.job_role = job_role
+        self.cv_summary = cv_summary
+        self.current_q = 0
+        
+        # Initialize models
+        self.embedder = SentenceTransformer('all-MiniLM-L6-v2')
+        self.llm = InferenceClient(model="meta-llama/Meta-Llama-3-8B-Instruct")
+        
+        # Setup FAISS index
+        self.index = IndexFlatIP(384)  # Embedding dimension
+        self._setup_knowledge_base()
+        
+        # Generate questions
+        self.questions = self._generate_questions()
+    
+    def _setup_knowledge_base(self):
+        """Load job requirements into vector store"""
+        reqs = {
+            "Software Engineer": [
+                "Explain SOLID principles",
+                "How would you optimize a slow SQL query?",
+                "Describe your experience with unit testing"
+            ],
+            "Data Scientist": [
+                "Explain bias-variance tradeoff",
+                "How would you handle missing data?",
+                "Describe a machine learning project you worked on"
+            ]
+        }
+        
+        self.knowledge = reqs[self.job_role]
+        embeddings = self.embedder.encode(self.knowledge)
+        self.index.add(embeddings)
+    
+    def _generate_questions(self):
+        """Generate questions using RAG"""
+        questions = []
+        
+        # Get most relevant requirements
+        cv_embed = self.embedder.encode(self.cv_summary["text"])
+        _, indices = self.index.search(np.array([cv_embed]), 3)
+        
+        for i in indices[0]:
+            prompt = f"""
+            Generate an interview question for a {self.job_role} position
+            based on this requirement: {self.knowledge[i]}
+            Candidate's CV summary: {self.cv_summary['text'][:1000]}
+            """
+            
+            response = self.llm.text_generation(
+                prompt,
+                max_new_tokens=100,
+                temperature=0.7
+            )
+            questions.append({
+                "text": response.strip(),
+                "base_requirement": self.knowledge[i]
+            })
+        
+        return questions
+    
+    def get_current_question(self):
+        return self.questions[self.current_q]
+    
+    def next_question(self):
+        self.current_q += 1
+    
+    def evaluate_answer(self, answer):
+        """Evaluate answer against knowledge base"""
+        # Get relevant requirement
+        req = self.questions[self.current_q]["base_requirement"]
+        
+        prompt = f"""
+        Evaluate this interview answer on a scale of 1-10:
+        Requirement: {req}
+        Answer: {answer}
+        
+        Provide JSON output with: score, feedback
+        """
+        
+        response = self.llm.text_generation(
+            prompt,
+            max_new_tokens=200,
+            temperature=0.3
+        )
+        
+        try:
+            return json.loads(response.split("{")[1].split("}")[0] + "}")
+        except:
+            return {"score": 5, "feedback": "Evaluation failed"}