Update src/streamlit_app.py

src/streamlit_app.py

@@ -36,26 +36,30 @@ from qa import retrieve_chunks, generate_answer, cache_embeddings, embed_chunks,
 # 🧠 LANGUAGE DETECTION HELPER (Improved for Hindi PDFs)
 # ==========================================================
 import re
+from langdetect import detect  # keep as fallback
 
 def detect_language(text_sample: str) -> str:
     """
-    Detect whether the document is primarily Hindi (Devanagari script) or English.
-    More reliable than langdetect for mixed or scanned PDFs.
+    Quick robust detection:
+    - If Devanagari chars present → Hindi (hi)
+    - Else fallback to langdetect (which needs real text to be accurate)
     """
     try:
-        # Count Devanagari and Latin characters
-        devanagari_chars = len(re.findall(r'[\u0900-\u097F]', text_sample))
-        latin_chars = len(re.findall(r'[A-Za-z]', text_sample))
-
-        # Decide dominant language
-        if devanagari_chars > latin_chars * 2:
+        # Fast deterministic check for Devanagari (Hindi) chars
+        if re.search(r"[\u0900-\u097F]", text_sample):
             return "hi"
-        else:
-            return "en"
+
+        # Some other Indic scripts? you can add more ranges similarly
+        # e.g. Bengali \u0980-\u09FF ; Tamil \u0B80-\u0BFF etc.
+
+        # Fallback to langdetect for everything else
+        lang = detect(text_sample)
+        return "hi" if lang.startswith("hi") else "en"
     except Exception:
         return "en"
 
 
+
 # ==========================================================
 # 🧠 SMART SUGGESTION GENERATOR
 # ==========================================================