---
title: "Discussion on \"Voice Agent Went From 66% to 96% Accuracy. Here's the Exact Optimization Loop I Used.\" | Hashnode"
description: "Discussion on \"Voice Agent Went From 66% to 96% Accuracy. Here's the Exact Optimization Loop I Used.\". How synthetic data generation, scenario simulation, and automated prompt optimization fixed a production voice agent without touching a single real user call."
url: "https://hashnode.com/posts/voice-agent-went-from-66-to-96-accuracy-here-s-the-exact-optimization-loop-i-used/69de8fe891716f3cfb6cfeb4"
---

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

Discussion

# [Voice Agent Went From 66% to 96% Accuracy. Here's the Exact Optimization Loop I Used.](https://litellmtechnicalbreakdown.hashnode.dev/voice-agent-went-from-66-to-96-accuracy-here-s-the-exact-optimization-loop-i-used?utm_source=hashnode&utm_medium=feed)

How synthetic data generation, scenario simulation, and automated prompt optimization fixed a production voice agent without touching a single real user call.

Apr 14 · 5 min read · litellmtechnicalbreakdown.hashnode.dev  
Read the article

-   [#llm](https://hashnode.com/tag/llm)
-   [#voice-ai](https://hashnode.com/tag/voice-ai)
-   [#promptengineering](https://hashnode.com/tag/promptengineering)
-   [#ai](https://hashnode.com/tag/ai)

## Written by

[jay](https://hashnode.com/@techwithj)

@techwithj

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

## Responses

Markdown supported · ⌘↵ to post

No responses yet.
