Update src/qa.py

src/qa.py

@@ -3,7 +3,7 @@ qa.py — GPT-4o (SAP Gen AI Hub) + ReRank Retrieval
 --------------------------------------------------
 ✅ Semantic retrieval (FAISS + cosine re-rank + neighbor fill)
 ✅ Bullet-aware similarity boost for procedural chunks
-✅ Embedding caching (per PDF)
+✅ Embedding caching (per PDF + chunk config aware)
 ✅ Smart factual mode (fast)
 ✅ Deep reasoning mode (ChatGPT-like)
 ✅ genai_generate() helper for suggestions
@@ -78,31 +78,35 @@ except Exception as e:
     chat_llm = None
 
 # ==========================================================
-# 4️⃣ Embedding Cache Manager
+# 4️⃣ Embedding Cache Manager (Chunk-Aware)
 # ==========================================================
 CACHE_EMB_DIR = "/tmp/embed_cache"
 os.makedirs(CACHE_EMB_DIR, exist_ok=True)
 
-def _hash_name(file_name: str):
-    """Generate unique hash for PDF file name."""
-    return hashlib.md5(file_name.encode()).hexdigest()
+def _hash_name(file_name: str, chunk_size: int, overlap: int, num_chunks: int):
+    """Generate unique hash based on file name and chunking config."""
+    combo = f"{file_name}_{chunk_size}_{overlap}_{num_chunks}"
+    return hashlib.md5(combo.encode()).hexdigest()[:8]
 
-def cache_embeddings(file_name: str, chunks, embed_func):
+def cache_embeddings(file_name: str, chunks, embed_func, chunk_size: int = None, overlap: int = None):
     """
-    Checks if cached embeddings exist for a PDF; if not, compute and save.
+    Load cached embeddings if available for the same document and configuration.
+    If not found, compute and store embeddings for reuse.
     """
-    cache_path = os.path.join(CACHE_EMB_DIR, f"{_hash_name(file_name)}.pkl")
+    cache_key = _hash_name(file_name, chunk_size or 1000, overlap or 100, len(chunks))
+    cache_file = f"{os.path.basename(file_name)}_cs{chunk_size}_ov{overlap}_{cache_key}.pkl"
+    cache_path = os.path.join(CACHE_EMB_DIR, cache_file)
 
     if os.path.exists(cache_path):
-        print(f"🧠 Loaded cached embeddings for {file_name}")
+        print(f"🧠 Loaded cached embeddings for {os.path.basename(file_name)} ({chunk_size}/{overlap})")
         with open(cache_path, "rb") as f:
             return pickle.load(f)
 
-    print(f"💡 No cache found for {file_name}. Generating embeddings...")
+    print(f"💡 No cache found for {os.path.basename(file_name)} ({chunk_size}/{overlap}). Generating new embeddings...")
     embeddings = embed_func(chunks)
     with open(cache_path, "wb") as f:
         pickle.dump(embeddings, f)
-    print(f"💾 Cached embeddings saved for {file_name}")
+    print(f"💾 Cached embeddings saved as {cache_file}")
     return embeddings
 
 def embed_chunks(chunks, batch_size=32):