Orchestrate Real-time Inference and Voice Interaction

Integrate RunPod for serving a lightweight, specialized ad rendering model (e.g., for generating specific visual styles or overlay text). Design the workflow so that after Gemini 2.5 Pro generates the core ad concept, RunPod refines its visual presentation. Additionally, integrate LiveKit to enable voice input for user preferences and voice output for ad delivery, ensuring a low-latency, engaging experience.