1) Ingest Files
PDF, DOC/DOCX/TXT, images (PNG/JPG), and audio (MP3/WAV/M4A/OGG/FLAC/WebM).
2) Build / Rebuild Index
Embeds text (incl. audio transcripts) with e5-small and writes a FAISS index.
3) Ask
Endpoint returns top-1 result. Response mirrors hit.text in summary.