Build RAG index at startup from transcripts; drop binary from git.

HF Spaces reject binary blobs in git history. Ship text transcripts only;
load_index builds data/index.npz on first use when missing.

.gitignore

@@ -1,6 +1,5 @@
-# Local data (generated); ship compact index + transcripts for Spaces
+# Local index is generated; ship transcripts only (Space builds index on startup)
 data/*
-!data/index.npz
 !data/transcripts/
 !data/transcripts/**
 

app.py

@@ -138,10 +138,9 @@ _warmup_error: str | None = None
 
 
 def _ensure_index():
-    if not Path(INDEX_PATH).exists():
-        raise FileNotFoundError(
-            f"Index not found at {INDEX_PATH}. Run `witgym-index` or bundle data/index.npz in the Space."
-        )
+    from witgym.retriever import load_index
+
+    load_index(INDEX_PATH)
 
 
 def _get_shared():

witgym/retriever.py

@@ -5,6 +5,7 @@ Returns analogous situations (same violation type), not similar words.
 """
 import json
 import os
+from pathlib import Path
 from typing import List
 from loguru import logger
 import numpy as np
@@ -27,6 +28,20 @@ def load_index(index_path: str = config.INDEX_PATH) -> dict:
         if os.path.exists(npz_fallback):
             path = npz_fallback
 
+    if not os.path.exists(path):
+        transcript_dir = Path(config.TRANSCRIPT_DIR)
+        if transcript_dir.is_dir() and any(transcript_dir.glob("*.txt")):
+            from witgym.indexer import build_index
+
+            build_path = path if path.endswith(".npz") else config.INDEX_PATH
+            logger.info(f"Index missing — building {build_path} from {transcript_dir}")
+            build_index(index_path=build_path)
+            path = build_path
+        else:
+            raise FileNotFoundError(
+                f"Index not found at {index_path} and no transcripts in {config.TRANSCRIPT_DIR}"
+            )
+
     if path.endswith(".npz"):
         archive = np.load(path, allow_pickle=True)
         scenes = [