Great breakdown of the architecture decisions. The monolith-first vs microservices question is especially relevant — premature microservices have killed more startups than monolith scaling issues.
We went monolith-first with AnveVoice (voice AI for websites) and it was the right call. Our voice agent handles real-time DOM actions — booking forms, navigation, search — and having everything in one process meant sub-700ms latency from day one. We only extracted the TTS pipeline into a separate service when it became a genuine bottleneck.
For booking engines specifically, the real-time availability check is where architecture matters most. Have you explored adding voice-based booking as an interaction layer? We're seeing travel sites add voice navigation for accessibility compliance (WCAG 2.1 AA) and it's becoming a competitive differentiator.
Solid thread — would love to hear what database strategy worked best for your concurrency-heavy booking scenarios.