Project
About project
About project
About project
The market offers many services for translating from one language to another, but none of them help you understand why a sentence is structured a certain way, why a particular verb form is used, or what an idiom actually means.
The Doubletapp ML team created a tool that not only translates but also helps you understand the language.
Our Telegram bot:
- translates text from Russian into a foreign language and back
- provides a detailed breakdown of each word and its role in the sentence
- explains grammar
- shows usage examples
- gives audio with correct pronunciation
- recognizes idioms and explains their meaning
Currently, the bot supports English, Spanish, German, Chinese, and Portuguese.
It’s useful for those who:
- are learning a language from scratch and want to understand the basics on their own
- are preparing for exams
- are traveling
- want to improve their listening skills
- are studying a language for work
So far, 2,280 users have tried the bot, and it has processed over 50,000 requests.
How the Bot Works
How the Bot Works
How the Bot Works
To use the bot, follow the link: t.me/perevod_smart_bot
Select a language and type the word or phrase you want to know the meaning of.
The bot will send you:
- a translation of the text into Russian
- a breakdown of the grammatical forms used in the text
- comments highlighting specific regional nuances — for example, Fahrenheit will be converted to Celsius
- additional useful phrases related to the topic
- an audio version of the phrase in the original language, to help you practice pronunciation.
How We’re Developing the Product
How We’re Developing the Product
How We’re Developing the Product
We are gradually adding new languages, planning potential integrations with other apps, and expanding the explanation capabilities so users can learn languages at a deeper level.
You can test the bot — just follow the link, choose a language, and start using it. Every day you can make 5 free requests.
We welcome feedback on what works well, what’s missing, and what would be great to improve.
Our super team
KirillBackend
IgorBackendImplementation
Implementation
Implementation
The bot’s functionality is powered by the GPT-4.1 mini model. We developed special prompts that define exactly how to process user translation requests. These prompts vary depending on the language pair — for example, Russian to English, Russian to Portuguese, and vice versa. Each prompt contains several examples to fine-tune the model’s output.
When a user sends a request, it is combined with the appropriate prompt and passed to the language model. In return, the bot provides a translation that takes into account the context and nuances of the language.
To reduce token costs, the prompt is placed at the beginning, and the user’s message follows it. This way, the LLM can apply prefix token caching (KV cache) to the repeating part of the prompt, making it 4× cheaper.
For convenience, after selecting a foreign language, the user can write either in Russian (the system will automatically translate into the foreign language) or in the foreign language (and the system will automatically translate into Russian). To detect which language the user’s message is in, the lingua-py library is used.
