How can I implement a vector search workflow that handles multiple data modalities, including text, images, and audio files? | Parse