fix: Correct data paths and embeddings fallback for production deployment

backend/config.py

@@ -31,10 +31,10 @@ class Settings(BaseSettings):
     port: int = int(os.getenv("PORT", "8000"))
     debug: bool = os.getenv("DEBUG", "true").lower() == "true"
     
-    # Data Storage
+    # Data Storage (paths relative to project root)
-    raw_news_dir: str = os.getenv("RAW_NEWS_DIR", "data/raw_news")
+    raw_news_dir: str = os.getenv("RAW_NEWS_DIR", "../data/raw_news")
-    processed_news_dir: str = os.getenv("PROCESSED_NEWS_DIR", "data/processed_news")
+    processed_news_dir: str = os.getenv("PROCESSED_NEWS_DIR", "../data/processed_news")
-    vector_index_path: str = os.getenv("VECTOR_INDEX_PATH", "data/news_vectors.faiss")
+    vector_index_path: str = os.getenv("VECTOR_INDEX_PATH", "../data/news_vectors.faiss")
     
     # Embedding Model
     embedding_model: str = "sentence-transformers/all-MiniLM-L6-v2"

backend/embeddings.py

@@ -155,10 +155,14 @@ class EmbeddingGenerator:
                 return np.array(response.embeddings[0])
             except Exception as e:
                 print(f"Cohere query embedding error: {e}")
-                # Fallback to sentence transformer
+                # Fallback to simple embeddings
-                return self.sentence_model.encode([query], convert_to_numpy=True)[0]
+                return self._simple_text_to_vector(query)
         else:
-            return self.sentence_model.encode([query], convert_to_numpy=True)[0]
+            if self.sentence_model is not None:
+                return self.sentence_model.encode([query], convert_to_numpy=True)[0]
+            else:
+                # Use simple hash-based embeddings
+                return self._simple_text_to_vector(query)
     
     def compute_similarity(self, embedding1: np.ndarray, embedding2: np.ndarray) -> float:
         """Compute cosine similarity between two embeddings"""