Monolingualists who want to communicate with global masses have never had it so easy. The old trusted Google translation can convert the content of images, audio and complete websites in hundreds of languages, while the newest tools such as Chatgpt also serve as useful pocket translators.
In the back, Deep and Oncenlabs have They have reached the high valuations of one billion dollars for various intelligent intelligent language related to the language that companies can channel to their own applications. But a new player is now entering the fray, with a location engine with AI that serves the infrastructure to help developers to globalize, a “strip” for the location of applications, if you wish.
Previously known as retropxication, Lingo.dev It is addressed to developers who wish to make the front of their application completely located from the beginning; All that they should worry is to send their code as usual, with lingo.dev bubbling under the hood on the autopilot. The result is that there is no copy/stuck text between chatgpt (for fast and dirty translations), or playing with multiple translation files in different formats obtained from innumerable agencies.
Today, Lingo.dev has clients such as French Unicorn Ai Mistral and Open Source Calendly Rival Cal.com. To boost the following phase of growth, the company has announced that it has raised $ 4.2 million in a round of initial funds led by initialized capital, with the participation of and Combinator and a series of angels.
Found in translation
Lingo.dev is the work of the CEO Max Prilutskiy and cpo Veronica Prilutskaya (in the photo above) who announced that they sold a previous startup of Saas called NOTION still Buyer not revealed last year. The duo had already been working in the foundations of Lingo.dev since 2023, with the first prototype developed as part of a Hackathon at Cornell University. This led its first clients who pay, before joining the Autumn and Combinator (YC) program last year.
In essence, Lingo-Dev is a translation API that developers can call locally Through its cli (Command line interface), or by direct integration with its CI/CD system through GITHUB or GITLAB. So, in essence, development equipment receives requests for extraction with automated translation updates every time a standard code change is made.
In the heart of all this, as expected, there is a large language model (LLM), or several LLM, to be exact, with lingo.dev orchestrating the various entries and exits among all of them. This mixing and combination approach, which combines anthropic models, OpenAi, among other suppliers, is designed to ensure that the best model for the task in question is chosen.
“The different indications work better in some models on other models,” Prilutskiy explained to TechCrunch. “Also depending on the case of use, we might want a better latency or latency may not matter everything.”
Of course, it is impossible to talk about LLM without also talking about the privacy of the data, one of the reasons why some companies They have been slower Adopt the generative AI. But with lingo.dev, the approach focuses substantial to locate the front interfaces, although it also serves commercial content, such as marketing sites, automated electronic emails and more, but does not channel the personal identification information of any client (PII (PII (PII ), For example.
“We do not expect personal data to be sent to us,” said Prilutskiy.
Through Lingo.dev, companies can build translation memories (a previously translated content store) and load their style guide to adapt the brand’s voice for different markets.
Companies can also specify rules on how particular phrases should be handled and in what situations. In addition, the engine can analyze the placement of the specific text, causing the necessary adjustments along the way, for example, a word when translated from English to German could have twice as much characters, which means that it would break the user interface. Users can instruct the engine to avoid this problem by reformulating a piece of text to match the length of the original text.
Without the broader context of what an application is really, it can be difficult to locate an independent piece of text, such as an interface label. Lingo.dev approaches this using a characteristic called “context awareness”, so it analyzes the entire content of the location file, including text keys or the adjacent event system that sometimes have translation files. It is about understanding the “microcontext”, as Prilutskiy expresses.
And it also comes more in this front in the future.
“We are already working on a new feature that uses screenshots of the application user interface, which Ingo.dev would use to extract even more contextual suggestions about the elements of the user interface and its intention,” he said.
Local go
It is still quite early for lingo.dev in terms of its path to full location. For example, colors and symbols can have different meanings between different cultures, something that Ingo.dev does not attend directly. In addition, things like metric/imperial conversions is something that the developer must still address at the code level.
However, Lingo.dev admits the Messageformat Marco, which manages the differences in the pluralization and the specific phrase of gender between languages. The company also recently launched an experimental beta function specifically for idioms; For example, “killing two birds with a stone” has an equivalent in German that translates approximately in “hitting two flies with a swat.”
In addition to that, Lingo.dev is also carrying out the applied research to improve several facets of the automated location process.
“One of the complex tasks we are currently working is to preserve female/male versions of nouns and verbs when translating between languages,” said Prilutskiy. “Different languages encode different amounts of information. For example, the word ‘teacher’ in English is neutral in gender, but in Spanish is “teacher“(Man) or”Teacher“(female). Ensure that these nuances are kept correctly are under our applied research efforts.”
Ultimately, the game plan is much more than a simple translation: it wants things to be as close as possible than you could get with a team of professional translators.
“In general, the [goal] With lingo.dev is to eliminate the friction of the location so thoroughly that it becomes a layer of infrastructure and natural part of the technological battery, ”said Prilutskiy. “Similar to how Strip eliminated the friction of online payments so effectively that it became a set of central developer tools for payments.”
While the founders more recently were headquarters in Barcelona, they are moving their formal home to San Francisco. The company counts only three employees, with a founding engineer who constitutes the trio, and this is a thin start philosophy that plan to follow.
“Friends of YC, me and other founders, we all believe in that,” said Prilutskiy.
His previous startup, which provided analysis for the notion, was completely torn with high profile clients, including Square, Shopify and Capital Sequoia, and had a large total of zero employees beyond Max and Veronica.
“We were two people, full -time, but with some contractors for several things from time to time,” added Prilutskiy. “But we know how to build things with minimal resources. Because the previous company was started, so we had to find a way for that to work. And we are replicating the same Lean style, but now with funds. “