refactored: to utilise the google adk and production grade agent

eval/eval_harness.py

@@ -1,27 +1,26 @@
-"""Evaluation Harness Runner for GCP Solution Architecture Agent."""
+"""Evaluation Harness Runner for GCP Solution Architecture Agent.
+
+Uses google.adk.evaluation.Evaluator with PostgreSQL database metrics persistence.
+"""
 
 import json
 import logging
 from pathlib import Path
 from typing import Any, Dict, List
 
+from app.adk.evaluation import ADKEvaluator
 from app.config import get_settings
-from app.skills.loader import SkillLoader
-from app.workflows.gcp_architecture_graph import create_agent
-from eval.metrics import evaluate_case_run
 
 logger = logging.getLogger(__name__)
 
 
 class EvalHarness:
-    """Offline Evaluation Harness for running benchmark suites against agent workflows."""
+    """Offline Evaluation Harness for running benchmark suites against ADK agents."""
 
     def __init__(self, dataset_path: Path | None = None) -> None:
         settings = get_settings()
         self.dataset_path = dataset_path or settings.EVAL_DATASET_PATH
-        self.skill_loader = SkillLoader(settings.SKILLS_DIR)
-        self.skill_loader.load_skills()
-        self.agent = create_agent(self.skill_loader)
+        self.evaluator = ADKEvaluator()
 
     def load_benchmark_cases(self) -> List[Dict[str, Any]]:
         """Load benchmark dataset JSON."""
@@ -33,7 +32,7 @@ class EvalHarness:
             return json.load(f)
 
     def run_eval_suite(self) -> Dict[str, Any]:
-        """Execute all benchmark test cases and compile scoring metrics."""
+        """Execute all benchmark test cases through ADKEvaluator and compile scoring metrics."""
         cases = self.load_benchmark_cases()
         if not cases:
             return {"status": "error", "message": "No benchmark cases loaded."}
@@ -42,15 +41,8 @@ class EvalHarness:
         total_passed = 0
 
         for case in cases:
-            logger.info("Evaluating benchmark case: %s", case.get("id"))
-            initial_state = {
-                "workflow_request": case.get("workflow_request", ""),
-                "target_dir": ".",
-                "active_skills": [],
-            }
-
-            final_state = self.agent.invoke(initial_state)
-            eval_result = evaluate_case_run(final_state, case)
+            logger.info("Evaluating ADK benchmark case: %s", case.get("id"))
+            eval_result = self.evaluator.evaluate_benchmark_case(case)
             results.append(eval_result)
 
             if eval_result.get("passed"):
@@ -73,7 +65,7 @@ def main() -> None:
     """CLI Runner for Evaluation Harness."""
     harness = EvalHarness()
     summary = harness.run_eval_suite()
-    print("=== GCP Solution Architecture Agent Benchmark Summary ===")
+    print("=== GCP Solution Architecture Agent ADK Benchmark Summary ===")
     print(json.dumps(summary, indent=2))