I've been frustrated with AI translation tools that produce inconsistent terminology and unnatural phrasing. So I built Jta, a CLI tool where the AI doesn't just translate—it critiques and improves its own work through "agentic reflection."
Instead of one-shot translation, Jta implements a 3-step cycle: 1. Translate 2. AI evaluates its own work (accuracy, fluency, style, terminology) 3. AI applies its own suggestions to improve
The trade-off: 3x API calls, but significantly better quality. For our production i18n files, this eliminated ~90% of manual fixes we used to do.
GitHub: https://github.com/hikanner/jta
.png)


