The development team working on the Menusso project made a change in the architecture just before the global premiere of the application.
We must distinguish that at this moment we are a step away from overthrowing the machine translation system.
This is a groundbreaking moment, and as you read this, you might almost yawn thinking it's another article about AI.
However, I must surprise you because the issue here is exciting from the perspective of the entire project.
Originally, Menusso was supposed to be revolutionary just from the aspect of relieving a restaurateur with machine translation of a given dish description and its ingredients. We assumed in 2022 that we would use Deeply, however, the launch of the Open AI API definitely changed our approach. It's super important to understand the meaning of local dish names and their subsequent translations into completely different languages, especially since in the prompt we input information about the type of cuisine and the fact that we are translating a meal description.
Marotino's In-Depth Look at DeepL and OpenAI APIs
Unfortunately, the Open AI API, as of the day I write this article, is not too fast. However, we transfer each translation to the database. So, when we want to call up exactly the same entry later, we ask the database and we don't lose the costs of re-translating. This way, we save time and translation costs. In Menusso, we give the possibility of editing the final translation result for the restaurant owner, I would call it with a tennis term 'angel touch', in case a human has the final say in this whole mess :)
Entering the year 2024, definitely, in terms of translations, we definitely recommend Open AI if we care about eliminating errors in translation logic. However, if we want to quickly translate single words, we can remain with DeepL, for example, as it will be faster and cheaper for the entire project.