Update app.py

app.py

@@ -59,8 +59,19 @@ def load_ohamlab_knowledge():
     return chunks
 
 # ---------------------------
-# 3. Generate or Load Embeddings (with Cache)
+# 3. Generate or Load Embeddings (with Cache & Retry)
 # ---------------------------
+def create_embeddings_with_retry(texts, retries=3, delay=2):
+    """Generate embeddings with retries on failure."""
+    for attempt in range(retries):
+        try:
+            res = client.embeddings.create(model=EMBED_MODEL, input=texts)
+            return [d.embedding for d in res.data]
+        except Exception as e:
+            print(f"⚠️ Embedding attempt {attempt+1} failed: {e}")
+            time.sleep(delay)
+    raise RuntimeError("❌ Failed to generate embeddings after retries.")
+
 def get_embeddings_with_cache():
     """Generate or load cached embeddings for OhamLab knowledge."""
     if os.path.exists(CACHE_PATH):
@@ -77,16 +88,12 @@ def get_embeddings_with_cache():
     chunks = load_ohamlab_knowledge()
     texts = [c["text"] for c in chunks]
     print(f"📘 Generating embeddings for {len(texts)} OhamLab chunks...")
+
     all_embs = []
     for i in range(0, len(texts), 50):
         batch = texts[i:i + 50]
-        try:
-            res = client.embeddings.create(model=EMBED_MODEL, input=batch)
-            embs = [d.embedding for d in res.data]
-            all_embs.extend(embs)
-        except Exception as e:
-            print(f"⚠️ Embedding batch failed ({i}): {e}")
-            all_embs.extend([[0.0] * 1536] * len(batch))  # fallback
+        embs = create_embeddings_with_retry(batch)
+        all_embs.extend(embs)
         time.sleep(0.5)
 
     data = [{"text": t, "embedding": e} for t, e in zip(texts, all_embs)]
@@ -103,7 +110,7 @@ OHAMLAB_TEXTS, OHAMLAB_EMBS = get_embeddings_with_cache()
 def retrieve_knowledge(query, top_k=3):
     """Retrieve top-k most relevant text snippets from markdown knowledge bank."""
     try:
-        q_emb = client.embeddings.create(model=EMBED_MODEL, input=[query]).data[0].embedding
+        q_emb = create_embeddings_with_retry([query])[0]
         sims = np.dot(OHAMLAB_EMBS, q_emb) / (
             np.linalg.norm(OHAMLAB_EMBS, axis=1) * np.linalg.norm(q_emb)
         )