Multimodal chatbot: orchestrating text, image, and audio

Workflow: 1) Session manager: Decide modality flows based on user input and device. 2) Text: Core dialog and tool calling. 3) Image: For uploads, run OCR and vision tagging; store lightweight fingerprints. 4) Audio: ASR for input; TTS for output with voice style controls. 5) Routing: Switch models by task; keep latency budgets per modality. 6) Safety: Per-modality moderation; image redaction; audio profanity filter. 7) Caching: Reuse embeddings and TTS segments. Delivers a cohesive experience across modalities within predictable SLAs.