update to the process_uploads upload.py and new database table called data

package/transcribe_video.py

@@ -12,17 +12,6 @@ os.environ["KMP_DUPLICATE_LIB_OK"] = "TRUE"
 # 5 minutes per chunk (300 seconds) keeps RAM usage low and stable
 CHUNK_DURATION_SEC = 300 
 
-print("🔧 Initializing C++ Engine...")
-model = WhisperModel(
-    "base", 
-    device="cpu", 
-    compute_type="int8", 
-    cpu_threads=0, # Let CTranslate2 auto-detect safe core counts
-    num_workers=1         
-)
-print("✅ C++ Model loaded successfully.")
-
-
 def extract_audio_and_chunk(video_path: str, output_dir: Path) -> list:
     """Extracts and splits audio into 5-minute chunks using a single FFmpeg pass."""
     print("🚀 Extracting and chunking audio with FFmpeg...")
@@ -49,6 +38,18 @@ def extract_audio_and_chunk(video_path: str, output_dir: Path) -> list:
 
 
 def transcribe_to_srt(video_path: str, force: bool = False):
+    if not hasattr(transcribe_to_srt, "model"):
+        print("🔧 Initializing C++ Engine...")
+        transcribe_to_srt.model = WhisperModel(
+            "base", 
+            device="cpu", 
+            compute_type="int8", 
+            cpu_threads=0, # Let CTranslate2 auto-detect safe core counts
+            num_workers=1         
+        )
+        print("✅ C++ Model loaded successfully.")
+
+
     video_path_obj = Path(video_path)
     srt_path = video_path_obj.with_suffix(".srt")
     temp_dir = video_path_obj.parent / f"temp_chunks_{video_path_obj.stem}"
@@ -78,7 +79,7 @@ def transcribe_to_srt(video_path: str, force: bool = False):
                 time_offset = chunk_idx * CHUNK_DURATION_SEC
                 print(f"\n⏳ Processing chunk {chunk_idx + 1}/{len(audio_chunks)} ({chunk_path.name})...")
                 
-                segments_generator, info = model.transcribe(
+                segments_generator, info = transcribe_to_srt.model.transcribe(
                     str(chunk_path), 
                     beam_size=1,
                     vad_filter=True,