Update app.py

- patched the answer function to showcase only the the top match’s text.

- Matches, still returns full top-k with scores for debugging.

- Cleaner, professional output

app.py

@@ -62,28 +62,25 @@ def answer(q, k, max_context_chars):
         return {"answer": "Index is empty. Ingest first.", "matches": []}
     qv = _normalize(_model.encode([q], convert_to_numpy=True))
     D, I = _index.search(qv, int(k))
+
     matches = []
     for i, s in zip(I[0].tolist(), D[0].tolist()):
-        if i < 0: continue
+        if i < 0:
+            continue
         matches.append({
             "id": _ids[i],
             "score": float(s),
             "text": _texts[i],
             "meta": _metas[i]
         })
-    # compose simple answer from contexts
-    blob, total = [], 0
-    for m in matches:
-        t = m["text"]; cut = min(len(t), max_context_chars - total)
-        if cut <= 0: break
-        blob.append(t[:cut]); total += cut
-        if total >= max_context_chars: break
-    if not blob:
+
+    if not matches:
         out = "No relevant context."
     else:
-        lines = [ln for ln in " ".join(blob).split(". ") if ln]
-        hits = [ln for ln in lines if any(tok in ln.lower() for tok in q.lower().split())] or lines[:2]
-        out = "Based on retrieved context:\n- " + "\n- ".join(hits[:4])
+        # 👇 only use the top match for the answer
+        top = matches[0]["text"]
+        out = f"Based on retrieved context:\n- {top}"
+
     return {"answer": out, "matches": matches}
 
 with gr.Blocks(title="RAG-as-a-Service") as demo: