Advice on tech stack for real-time GPT scoring + spectator voting

I’m prototyping a real-time web app that uses GPT to evaluate text-based exchanges between two users while spectators react and vote.

For the MVP, I’m trying to figure out the best stack for:
– Low-latency chat with spectator interactions (likely WebSockets or Firebase)
– GPT scoring that ignores participant identities
– Fast verdict summaries with minimal API lag

Any recommendations or lessons learned from similar real-time + GPT projects?