PARTNERS & INTEGRATORS

No. TAUS EPIC uses dedicated Quality Estimation models. Unlike generic LLMs, our models are trained specifically on translation data to provide a reliable, confidence score without needing a human reference text.

We offer customization. If you have at least 25,000 translated segments, we can train a hyper-targeted model entirely on your corporate data and glossaries. This ensures the AI mimics your internal quality standards perfectly, and the resulting model is exclusively owned and accessed by you.

No. You are in control of the "threshold". You decide what score constitutes a "good" translation. Segments that meet your standard can be automatically approved, and anything below that threshold is flagged for your human editors to review.

The most effective way to maximize your integration is by upgrading from a generic model to a Customized QE Model. While generic models are great for standard use cases, a custom model is trained entirely on your corporate data. This means the AI learns exactly how your internal linguists detect and fix errors, making the quality scores hyper-targeted and highly relevant to your specific brand, industry, and quality standards.

No. Once your custom model is built, it automatically links to your existing API key and language pairs. Your integration—whether in Trados, Phrase, Crowdin, MemoQ, or Blackbird—will instantly start using the custom model without you needing to change any configurations or write new code.

There are two other key ways to calibrate your system:

• Find your threshold "sweet spot": Test a representative sample of your content to find the exact score that separates flawless translations from those needing review. For many, the sweet spot is around 0.86, but you choose this threshold based on your specific risk tolerance.

• Use RAG for automatic corrections: If you also use APE to fix poor segments, you can upload your translation memories and glossaries. The system uses Retrieval-Augmented Generation (RAG) to reference these files on the fly, ensuring the AI fixes errors using your approved brand terminology.