Monoline Twists who want to communicate with the global masses have never had it that easy. While the old, reliable Google Translate can convert images, audio, and content across your website in hundreds of languages, new tools like ChatGPT can also serve as a handy pocket translator.
On the backend, Deepl and 11Enlabs have reached a $1 billion valuation for various language-related smarts that businesses can pour into their own applications. But now, new players are in the fight. AI-powered localization engines serve infrastructure that help developers become global.
Formerly known as Replexica, Lingo.Dev is targeting developers who want to fully localize their app’s frontend from Get-go. All they need to worry is to ship the code as normal, and Lingo.dev will bubble under the hood of the autopilot. The result is that there is no copy/paste text between ChatGpt (for quick and dirty translations), or messing around with multiple translation files in different formats sourced from countless institutions.
Today, Lingo.Dev counts customers such as French Unicorn Mistral AI and open source calendar-ry rival Cal.com. To drive the next phase of growth, the company announced that it has raised $4.2 million in funding seed rounds led by initialized capital, with participation from Y-combinators and many angels.
Found in translation
Lingo.dev is hand-crafted by CEOs Max Prilutskiy and CPO Veronica Prilutskaya (pictured above), and announced last year that it had sold a previous SaaS startup called Notionlytics to private buyers. The duo has already been working on the fundamentals of Lingo.dev since 2023, with the first prototype being developed as part of the Cornell University hackathon. This led to first paying customers before participating in last year’s Y Combinator (YC) fall program.
In its core, Lingo-Dev is a translation API either invoked locally by the developer via the CLI (command line interface) or via direct integration with CI/CD systems via GitHub or GitLab. So essentially, the development team receives a pull request with automated translation updates whenever a standard code change is made.
As you can imagine, at the heart of all of these, you’ll be precisely a large-scale language model (LLM) or some LLM, and Lingo.Dev coordinates the various inputs and outputs between them all. This mix-and-match approach, combining human models, is designed to combine OpenAI models, among other providers, to select the best model for the task at hand.
“The different prompts work better on some models than on others,” Prilutskiy explained to TechCrunch. “In some use cases, you may need better latency, but everything may not be important.”
Of course, it is impossible to talk about LLM without even talking about data privacy. This is one reason why some companies are slow to adopt generative AI. However, Lingo.dev also supports business content such as marketing sites, automated email, and while it focuses essentially on the local front-end interface, it focuses on customer personal identifiable information. Does not (PII), for example.
“I don’t think that any personal data will be sent,” Prilutskiy said.
Through Lingo.dev, companies can build translated memory (stores of previously translated content), upload style guides, and adjust brand audio for different markets.
Companies can also specify how to handle a particular phrase and in any circumstances the rules are specified. Additionally, the engine can analyze the specific text arrangement and make the necessary adjustments along the way. For example, words when translated from English to German could double the number of characters that mean breaking the UI. The user can instruct the engine to rephrase the text to avoid that issue by matching the length of the original text.
Without a broader context about what an application actually is, it is difficult to localize small standalone text, such as labels on an interface. Lingo.dev works around this using a feature called “Context Awareness”. This analyses the entire content of a localization file that contains adjacent text or event system keys that the translation file sometimes has. As Prilutskiy says, it’s all about understanding “microcontext”.
And there’s more to come in this aspect in the future.
“We’re already working on a new feature that uses screenshots of the app’s UI. Lingo.dev is used to extract more contextual hints about UI elements and their intent.” he said.
Go local
For Lingo.dev, it’s still quite early in terms of paths to full localization. For example, colors and symbols can have different meanings across different cultures. This is something Lingo.dev does not directly support. What’s more, things like metric/imperial conversions are something that developers still need to deal with at the code level.
However, Lingo.dev supports the MessageFormat framework that handles differences between plural and gender-specific phrases between languages. The company recently released an experimental beta feature exclusively for idioms. For example, “killing two birds with one stone” is the German equivalent, translated into “killing two flys with one SWAT.”
In addition, Lingo.dev is also conducting applied AI research to improve various aspects of the automated localization process.
“One of the complicated tasks we’re working on now is to store feminine/masculine versions of nouns and verbs when translating between languages,” says Prilutskiy. “In different languages, we encode different amounts of information. For example, the word “teacher” in English is gender neutral, but in Spanish it is either “maestro” (male) or “mestra” (female). . Ensuring that these nuances are properly preserved is an applicable AI research effort. ”
Ultimately, the game plan is more than just a simple translation. Just like a team of professional translators, you want to get things as close as possible.
“Overall, the (goal) with Lingo.dev is to eliminate friction from localization very thoroughly, making it a natural part of the infrastructure layer and high-tech stack,” Prilutskiy said. Masu. “Stripe has effectively eliminated friction from online payments, making it a core developer toolkit for payments.”
The founder was based in Barcelona recently, but they have moved their official home to San Francisco. The company only counts three employees in total, with founding engineers making up the trio. This is a lean startup philosophy that you plan to follow.
“YC, me, and the other founders, we are all big fans of that,” Prilutskiy said.
Previous startups that provided analysis for the concept were completely bootstrapped with well-known customers including Square, Shopify and Sequoia Capital, with total zero employees exceeding Max and Veronica.
“We were two full-time, but sometimes we had contractors for different things,” added Prilutskiy. “But we know how to build things with minimal resources. The previous company was bootstrapped so we had to find a way that would work. And we were the same lean It replicates the style, but now it’s funded.”