We used OpenAI's new Realtime Voice API for the voice calling features, and it was a blast. We had also considered transcribing the user's voice requests, sending them to a text-based LLM and then using TTS to convert them to "Santa Voice", but no alternatives were as good as the Realtime API.
We used the RevenueCat REST API to manage and validate subscriptions in our backend. It was specially practical considering we didn't need to do our own receipt validation.