feat: QA self-check agent — 15 prompts, 5 judges, weighted scoring

scripts/qa/judges/gdino.py

@@ -0,0 +1,38 @@
+"""Grounding DINO judge: zero-shot object detection from prompt keywords"""
+import requests, json, io
+from PIL import Image
+
+GDINO_URL = "http://localhost:5051/search"
+DEFAULT_UUID = "aeed71342a899fe4b4c57b7d41bcb692"
+
+def score(frames, prompt):
+    prompt_lower = prompt.lower()
+    
+    # Just do a single time-bounded search (not per frame)
+    try:
+        resp = requests.post(GDINO_URL, json={
+            "file_uuid": DEFAULT_UUID,
+            "text": prompt_lower,
+            "limit": 3,
+            "start_time": 0,
+            "end_time": 0
+        }, timeout=30)
+        data = resp.json()
+        hits = data.get("hits", [])
+        n_hits = len(hits)
+        best_score = max((h.get("best_score", 0) for h in hits), default=0)
+        dets_found = []
+        for h in hits:
+            for d in h.get("detections", []):
+                dets_found.append(d.get("label", ""))
+        
+        score_val = int(100 * min(1.0, best_score * 2))
+        
+        return {
+            "agent": "GroundingDINO",
+            "score": score_val,
+            "reasoning": f"{n_hits} hits, best_score={best_score:.2f}, labels={dets_found[:3]}",
+            "details": {"n_hits": n_hits, "best_score": best_score}
+        }
+    except Exception as e:
+        return {"agent": "GroundingDINO", "score": 50, "reasoning": f"GDINO error: {str(e)[:80]}", "details": {}}