---
title: "Discussion on \"Why Your App Works in the Demo and Falls Over Under Load\" | Hashnode"
description: "Discussion on \"Why Your App Works in the Demo and Falls Over Under Load\". How we cut voice response time from 900ms to 320ms without touching the prompt"
url: "https://hashnode.com/posts/app-works-in-demo-falls-over-under-load/6ac752f01a129717f25e98e1"
---

[Feed](https://hashnode.com/feed) /

Discussion

# [Why Your App Works in the Demo and Falls Over Under Load](https://keencraft.hashnode.dev/app-works-in-demo-falls-over-under-load?utm_source=hashnode&utm_medium=feed)

How we cut voice response time from 900ms to 320ms without touching the prompt

Oct 8 · 6 min read · keencraft.hashnode.dev  
Read the article

-   [#backend](https://hashnode.com/tag/backend)
-   [#performance](https://hashnode.com/tag/performance)
-   [#postgresql](https://hashnode.com/tag/postgresql)
-   [#redis](https://hashnode.com/tag/redis)
-   [#system-design](https://hashnode.com/tag/system-design)

## Written by

[Muhammad Jareer](https://hashnode.com/@keencraft)

@keencraft · San Francisco, USA

  
[View profile](https://hashnode.com/@keencraft)

Real-time voice AI, LLM integrations and automation | Founder, KeenCraft

I build AI voice agents that book real appointments, plus custom software and automation, at KeenCraft. Lead developer on VoiceCake, Rawk.ai, and Dynaris, where I cut voice response latency from 900ms to 320ms. I write about voice AI architecture, latency, and backend systems: Python, FastAPI, LiveKit, Twilio, Postgres, and Redis.

## Responses

Markdown supported · ⌘↵ to post

No responses yet.
