Hi everyone. I'm trying to run a fairly typical voice conversation system using Twilio and OpenAI realtime API via websockets (using Node fastify) and Redis. This ran well on my local PC for testing with Ngrok, but I'm hoping deploying on Railway will give super smooth websockets and responses.

I'm wondering if my application is right for Railway environment.

The first test, using the provided domain/url partially worked. OpenAI could hear the media stream from the caller, but the caller could not hear what OpenAI sent back.

Things improved by using my own domain/url and the conversation within the call worked well. However in more than half the test calls, the caller doesn't hear the greeting, and towards the end of the call, the caller no longer hears the OpenAI responses.

Calls are about 4-7 mintues long and each turn of conversation is between 3 seconds and, perhaps 60 seconds of speaking.

I'm assuming these are related to websockets response times. Are there ways to improve the websocket response times in railway?

Project id is 0e6a359f-9099-49db-ab98-32669e6eb67a

Thanks!