It has never been so easy for a monolingualist who wants to communicate with the global masses. The trustworthy old Google Translate can convert images, audio and content across the entire website in hundreds of languages, and new tools like ChatGpt can also be used as a convenient pocket translator.

On the backend, DEEPL and ElevenLabs have added sublime valuations to various language-related intelligence that businesses can transfer to their own applications. But a new player is now entering the competition and adopting an AI-powered localization engine that serves infrastructure to help developers get to the world, which is the "stripe" of application localization, if you will.

Lingo.Dev was formerly known as Replexica. All they need to worry about is shipping the code as usual, lingo.dev bubbling under the hood on the autopilot. The result is that there is no copy/paste text between chatgpt (for quick and dirty translations), nor multiple translation files in different formats sourced from various agents.

Today, Lingo.dev calculates customers such as French unicorn Mistral AI and open source calendar rival Cal.com. To drive the next phase of growth, the company announced it has raised $4.2 million, a seed round led by Initial Capital, with the participation of Y Combinator and many angels.

Discovered in translation

Lingo.dev, a handicraft for CEO Max Prilutskiy and CPO Veronica Prilutskaya (pictured above), announced that they sold their former SaaS startup last year to a startup called Notionlytics, last year to an unsold one. Public buyer. The duo has been working on lingo.dev since 2023, and the first prototype was part of a hackathon at Cornell University. This led to their first paid customer and then joined Y Combinator (YC)’s fall program last year.

Lingo-dev is a translation API that can be called locally through its CLI (command line interface) through developers, or directly integrated with the CI/CD system through GitHub or GitLab. So essentially, the development team receives pull requests with automatic translation updates when making standard code changes.

As you would expect, the core of all this is a large language model (LLM) or several LLMs, and LINGO.DEV curates the various inputs and outputs between them. This mix and match method combines anthropomorphic OpenAI and other providers’ models to ensure the best model is selected for the task at hand.

"In some models, different hints work better in some models," Prilutskiy explained to TechCrunch. "Again, depending on the use case, we might want better latency, otherwise the latency may not matter. ”

Of course, it is impossible to talk about LLM without talking about data privacy, which is one of the reasons why some businesses adopt AI generation. But with the help of lingo.dev, the focus is on localized front-end interfaces, and although it also conforms to business content such as marketing websites, automatic emails, etc., it does not aggregate personal identity information (PIII) to any customer, for example.

"We do not want to send any personal data to us," Prilutskiy said.

Through Lingo.dev, companies can build translation memories (the store of previously translated content) and upload their style guides to customize brand voices in different markets.

Businesses can also specify rules about how a particular phrase should be processed and under what circumstances. Additionally, the engine can analyze the placement of specific text, making the necessary adjustments in the process - for example, the word translated to English to German may be twice as many characters as the number of characters, meaning it will destroy the UI. The user can bypass the problem by redrawing a piece of text to instruct the engine to match the length of the original text.

Without the practical situation of a wider application, it is difficult to locate a small portion of independent text (such as tags on the interface). lingo.dev solves this problem using a feature called "context awareness" to analyze the entire content of a localized file, including adjacent text or event system keys that the translated file sometimes has. As Prilutskiy said, this is all about understanding the "micro".

There will be more fronts in the future.

"We are already using the new feature of screenshots of the app UI, which Lingo.dev will use to extract more contextual hints about UI elements and their intent," he said.

Go local

As far as its path to full localization is concerned, Lingo.dev is still quite early. For example, colors and symbols may have different meanings between cultures, which lingo.dev does not cater directly. Additionally, developers still need to address things like metric/imperial conversion at the code level.

However, Lingo.dev does support the MessageFormat framework, which handles differences in diversity and gender-specific wording between languages. The company also recently released an experimental beta feature specifically for idioms; for example, the German equivalent of “kill two birds with a stone” roughly translated to “hit two flies with a SWAT.”

Most importantly, Lingo.dev is also conducting applied AI research to improve all aspects of the automatic positioning process.

"One of the complex tasks we are currently working on is to translate the female/male version of nouns and verbs between languages," Prilutskiy said. "Different languages ​​encode different amounts of information. For example, the word "teacher" in English is Gender neutral, but in Spanish it is”master” (male) orteacher"(Women). Make sure to keep these nuances right under the AI ​​research efforts we apply."

Ultimately, the game plan is much more than simple translation: it wants to make things as close as possible to what you can get with a professional translation team.

"Overall, the (target) of using Lingo.dev is so thoroughly eliminating the friction of localization that it becomes the infrastructure layer and natural part of the technology stack," Prilutskiy said. "Similar to how stripes eliminate it so effectively The friction in online payments, so much so that it becomes the core developer toolkit for payments.”

Although the founders recently reside in Barcelona, ​​they are moving their official home to San Francisco. The company has only three employees in total, with one of the founding engineers forming a trio, a lean entrepreneurial philosophy they plan to follow.

"YC, myself and the rest of the founders, we are all loyal believers in this," Pritski said.

Their previous startups provide analytics for the concept, guided entirely by high-profile clients including Square, Shopify and Sequoia Capital – and its total number exceeds Max and Veronica.

"We are two people, full-time, but now and then there are some contractors," Prilutskiy added. "But we know how to build things with minimal resources. Because previous companies were launched, we had to find a way." Come work. We are replicating the same lean style – but now there is funding.”