Tuned build_prompt further

agents/scorer.py

@@ -2,7 +2,6 @@
 agents/scorer.py
 ────────────────────────────────────────────────────────────────────────────
 ScorerAgent: Semantic, context-aware answer scoring — implements Agenda Item #4.
-Upgraded to leverage Qwen2.5-32B-Instruct for advanced conceptual grading.
 """
 
 import json
@@ -58,30 +57,32 @@ class ScorerAgent:
             "star_hint":      star_hint,
         }
 
-    # ── Advanced Prompt Builder ───────────────────────────────────────────────
+
+     # ── Advanced Prompt Builder ───────────────────────────────────────────────
     def _build_prompt(self, answer: str, question: str, industry: str, 
                       role_level: str, keywords: list) -> str:
         keywords_str = ", ".join(keywords) if keywords else "General Industry Domain"
         
-        return f"""You are an elite corporate technical interviewer. Evaluate the candidate's response using the STAR (Situation, Task, Action, Result) methodology.
+        return f"""You are an elite technical interviewer. Evaluate the candidate's response using the STAR methodology.
 
-Interview Question Asked: {question}
+Interview Question: {question}
 Candidate's Answer: {answer}
 Target Domain Keywords: {keywords_str}
 
-CRITICAL KEYWORD EVALUATION RULE:
-Do NOT run a rigid character match. If the candidate naturally explains the underlying concept, uses functional synonyms, describes the operational setup, or discusses the exact methodology related to a keyword, classify it as a SUCCESSFUL semantic match.
+🚨 CRITICAL KEYWORD EVALUATION RULES (READ CAREFULLY):
+1. Do NOT look for exact text characters or verbatim matches.
+2. If the candidate describes the concept of a keyword, mentions a synonym, or explains the real-world task/workflow related to that keyword, you MUST count it as a SUCCESSFUL match.
+3. For example: If the keyword is "Database", and they discuss "storing text files" or "SQL optimization", that is a successful match. Be highly generous with conceptual matches!
 
 Provide your evaluation in a raw, clean JSON format. Do not include any markdown block formatting (like ```json). Use exactly this structure:
 {{
-  "score": [Integer from 1 to 10 based on relevance and quality],
+  "score": [Integer from 1 to 10 based on content quality],
   "hit_keywords": [Array of string keywords from the Target list that were covered or conceptually explained],
   "missed_keywords": [Array of string keywords from the Target list completely missing in both text and concept],
   "strength": "One clear sentence highlighting what the candidate did well.",
   "weakness": "One clear sentence identifying the biggest structural or domain gap.",
   "fix": "One clear sentence containing actionable advice to improve the response."
 }}"""
-
     # ── Robust Safe JSON Parser ───────────────────────────────────────────────
     def _parse_json_response(self, raw_text: str, original_keywords: list) -> dict:
         """Safely parses LLM output and provides bulletproof defaults if parsing fails."""