diff --git a/wwwroot/knowledge_bases_list/upload_file.dspy b/wwwroot/knowledge_bases_list/upload_file.dspy
index 91c1add..5bebc4a 100644
--- a/wwwroot/knowledge_bases_list/upload_file.dspy
+++ b/wwwroot/knowledge_bases_list/upload_file.dspy
@@ -106,7 +106,10 @@ async def ingest_doc(doc_id, kb_id, file_name, ext_l, real_path):
                 fd = json.loads(resp)
                 results = fd.get("results", [])
                 if results and isinstance(results[0], dict):
-                    face_count = len(results[0].get("faces", results[0].get("detections", [])))
+                    faces = results[0].get("faces", results[0].get("detections", []))
+                    face_count = len(faces)
+                    if faces and isinstance(faces[0], dict):
+                        meta_parts['face_bboxes'] = [f.get("bbox", {}) for f in faces[:10]]
                 meta_parts['face'] = face_count
             except: pass
 
@@ -133,13 +136,16 @@ async def ingest_doc(doc_id, kb_id, file_name, ext_l, real_path):
                     with open(tmp_img, 'rb') as fi:
                         frame_data = fi.read()
                     img_b64 = base64.b64encode(frame_data).decode()
+                    frame_bboxes = []
                     try:
                         client = StreamHttpClient()
                         resp = await client.request('POST', 'https://media.opencomputing.net:10443/face/api/detect', json={"images": [img_b64]})
                         fd = json.loads(resp)
                         results = fd.get("results", [])
                         if results and isinstance(results[0], dict):
-                            face_count = len(results[0].get("faces", results[0].get("detections", [])))
+                            faces = results[0].get("faces", results[0].get("detections", []))
+                            face_count = len(faces)
+                            frame_bboxes = [f.get("bbox", {}) for f in faces[:10]] if faces else []
                     except: pass
                     # --- CLIP image embedding for video frame ---
                     try:
@@ -158,12 +164,16 @@ async def ingest_doc(doc_id, kb_id, file_name, ext_l, real_path):
                                 {"id": doc_id + "_c0", "vector": img_embeddings[0], "text": file_name}
                             ]}
                             await client3.request('POST', 'https://vectordb.opencomputing.net:10443/v1/upsert', json=vdb_data)
+                            chunk_meta = {"start_time": 0}
+                            if frame_bboxes:
+                                chunk_meta["bboxes"] = frame_bboxes
                             async with db.sqlorContext('rag') as sor:
                                 await sor.sqlExe(
-                                    "INSERT INTO document_chunks (id, doc_id, kb_id, chunk_index, content, vector_id, created_at) "
-                                    "VALUES (${id}$, ${doc_id}$, ${kb_id}$, 0, ${content}$, ${vid}$, NOW())",
+                                    "INSERT INTO document_chunks (id, doc_id, kb_id, chunk_index, content, vector_id, metadata, created_at) "
+                                    "VALUES (${id}$, ${doc_id}$, ${kb_id}$, 0, ${content}$, ${vid}$, ${meta}$, NOW())",
                                     {"id": doc_id + "_c0", "doc_id": doc_id, "kb_id": kb_id,
-                                     "content": file_name, "vid": doc_id + "_c0"})
+                                     "content": file_name, "vid": doc_id + "_c0",
+                                     "meta": json.dumps(chunk_meta, ensure_ascii=False)})
                         except:
                             pass
                     os.remove(tmp_img)