Post-Submission Update: The 5-Day Latency Battle.
We submitted Falah.ai to the hackathon about 30 minutes before the deadline, deployed on GCP Cloud Run. The submission was in, but the real-time agent had latency issues that were killing the conversational experience.
Since we couldn't push any code changes after the deadline, we spent the next 3–5 days purely tweaking GCP deployment configurations. Tried everything- different instance settings, scaling configs, region changes, you name it. I even brought in 2 other people to help troubleshoot. None of us were GCP experts honestly, but we gave it everything we had. Nothing worked.
Maybe a Google Cloud expert could have cracked it, but we didn't have one.
But I didn't lose hope. After days of hitting the same wall, I got a random idea what if the problem isn't my code at all? What if it's the deployment platform? So I decided to just verify: I deployed the exact same code, same main branch, not a single line changed onto Railway. Just to test.
And I was genuinely shocked. It worked perfectly. Near-zero latency. The agent was responding in real-time, smooth, natural exactly how we designed it. I'm attaching a video as proof, and our repo is public so you can check the last code push on the main branch yourself nothing changed. The code was never the problem. Same Gemini 2.5 Flash model, same LiveKit pipeline, same everything. Falah.ai is now live on both platforms, and the experience on Railway is exactly what we envisioned when we built this.
I damn sure if we setup right configs in GCP it will also work without any latency!
Video:https://youtu.be/SVJZTXPNGF0
Railway deployed link:https://lucky-harmony-production-8c2e.up.railway.app/
GCP deployed:https://frontend-falah-ai-310392085947.asia-south1.run.app/login
Log in or sign up for Devpost to join the conversation.