Since 2024, Caucasus Translations has supported an international online marketplace in Georgian, Azerbaijani and Armenian. The partnership is ongoing. Our in-house team in Tbilisi handled Georgian localisation for buyer and seller interfaces, tested the website and mobile apps, and evaluated how well machine translation systems handled Georgian e-commerce content.
- Engagement
- Ongoing since 2024
- Languages
- 3 working languages
- Georgian testing
- 79 hours in one Georgian testing round
From existing translations to a usable marketplace
The project began with a practical question: which existing Georgian translations could be edited, and which needed to be redone? Our Tbilisi team reviewed the existing copy before taking on new website, seller portal and mobile app content. One seller portal assignment covered more than 52,000 weighted words for translation, alongside a separate editing scope.
Our linguists maintained shared general and legal glossaries and reviewed automated QA alerts before delivery. In the interface, they checked how Georgian wording fit on screen and whether changing values, such as the number of items in a cart, would still display correctly.
What our Tbilisi team found in actual builds
Our in-house linguists and engineers spent 79 hours testing Georgian across web, iOS and Android in one completed round. On the website, a registration field rejected a standard Georgian mobile number. On Android, a sign-in notification appeared in another language even though the app was set to Georgian.
The team logged each issue by platform and screen, attached screenshots and supplied corrected wording where applicable. The marketplace’s engineers could see what happened, where it happened and what to check in the next build.
How we tested machine translation for Georgian
Our linguists compared the Georgian output of six machine translation engines, looking for changes in meaning, missing information, grammatical errors and unnatural wording. They recorded issues string by string and summarised the findings for each engine.
We also prepared human reference translations for product listings and evaluated Georgian output from a separate large language model. The project team received examples showing where the model’s output needed human review and what a usable Georgian version looked like.
Need to localise AI prompts or test an AI product in the South Caucasus?
For projects in Georgian, Azerbaijani or Armenian, tell us what your users ask the system to do and what a good answer looks like. We can scope the language work around the areas you need to test.
- AI prompt translation and localisation
- Adapted instructions, example inputs and conversation flows that preserve the intended task and sound natural in the target language.
- Multilingual chatbot or LLM testing
- Test scenarios and human review of responses for meaning, terminology, tone and whether the model followed the instructions.
- AI evaluation data and annotation
- Human-written reference translations or answers, plus annotated examples showing where output needs correction and why.
- Machine translation evaluation or post-editing
- Side-by-side model comparisons, an issue report, or human-edited copy ready for the agreed use.
- In-product testing and team handover
- Screenshots and issue reports from test builds, with a walkthrough of how your team can maintain terminology and retest fixes after updates.
Share a sample of your prompts, interface strings or model outputs, along with the target languages and release schedule. That gives us enough context to propose a pilot and agree on what to measure before you scale the work.
Request a project quote