Update app.py

app.py

@@ -69,12 +69,46 @@ class Pair:
     title: Optional[str]
     score: float
     image_path: Optional[str]
+    text: Optional[str] = None  # <-- NEW
+
+
+def _get_meta_text(m: dict) -> Optional[str]:
+    # Try common keys first
+    for k in ("text", "content", "passage", "body", "chunk", "article"):
+        if m.get(k):
+            return m[k]
+    # If you stored a local file path for the text, read it
+    p = m.get("path") or m.get("filepath")
+    if p:
+        import os
+        fp = p if os.path.isabs(p) else os.path.join(DATA_DIR, p)
+        if os.path.exists(fp):
+            try:
+                with open(fp, "r", encoding="utf-8") as f:
+                    return f.read()
+            except:
+                pass
+    return None
 
 def _pair_from_idx(idx: int, score: float, rank: int) -> Pair:
     m = TEXT_META[idx]
     img_path = IMAGE_META[idx].get("image_path")
-    return Pair(rank=rank, idx=idx, doc_id=m.get("id"), title=m.get("title"), score=float(score), image_path=img_path)
-
+    return Pair(
+        rank=rank,
+        idx=idx,
+        doc_id=m.get("id"),
+        title=m.get("title"),
+        score=float(score),
+        image_path=img_path,
+        text=_get_meta_text(m),  # <-- NEW
+    )
+
+def _truncate(s: str, max_chars: int = 1200) -> str:
+    if not s: return ""
+    s = s.strip().replace("\r", " ")
+    return s[:max_chars]
+
+    
 def search_text(q: str, topk: int = 10) -> List[Pair]:
     qv = text_enc.encode([q], convert_to_numpy=True, normalize_embeddings=True).astype("float32")
     D, I = T_INDEX.search(qv, topk)
@@ -97,12 +131,18 @@ def search_image(img: Image.Image, topk: int = 10) -> List[Pair]:
     return out
 
 def build_prompt(question: str, ctx: List[Pair]) -> str:
-    lines = ["You are a helpful assistant. Answer the user's question using the given context.",
-             "If the answer is not contained in the context, say you don't know.\n",
-             "Context:"]
+    lines = [
+        "از زمینهٔ زیر استفاده کن و به فارسی پاسخ بده. اگر پاسخ در زمینه نبود، بگو «نمی‌دانم».",
+        "",
+        "### زمینه:",
+    ]
     for p in ctx:
-        lines.append(f"- {p.title or ''} (id={p.doc_id}) [score={p.score:.3f}]")
-    lines.append(f"\nQuestion: {question}\nAnswer:")
+        snippet = _truncate(p.text or "")
+        lines.append(
+            f"- عنوان: {p.title or '—'} (id={p.doc_id}, score={p.score:.3f})\n"
+            f"  متن: {snippet if snippet else '—'}"
+        )
+    lines.append(f"\n### پرسش: {question}\n### پاسخ:")
     return "\n".join(lines)
 
 def call_llm(prompt: str) -> str: