Monolinguists wanting to communicate with the global masses have never had it so easy. Trusty old Google Translate can convert the content of images, audio, and entire websites across hundreds of languages, while newer tools such as ChatGPT also serve as handy pocket translators.
On the back end, DeepL and ElevenLabs have have reached lofty billion-dollar valuations for various language-related smarts that businesses can funnel into their own applications. But a new player is now entering the fray, with an AI-powered localization engine that serves the infrastructure to help developers go global — a “Stripe” for app localization, if you will.
Formerly known as Replexica, Lingo.dev targets developers who want to make their app’s front end fully localized from the get-go; all they need to worry about is shipping their code as usual, with Lingo.dev bubbling away under the hood on autopilot. The upshot is that there is no copy/pasting text between ChatGPT (for quick and dirty translations), or messing around with multiple translation files in different formats sourced from myriad agencies.
Today, Lingo.dev counts customers such as French unicorn Mistral AI and open source Calendly rival Cal.com. To drive the next phase of growth, the company has announced it has raised $4.2 million in a seed round of funding led by Initialized Capital, with participation from Y Combinator and a slew of angels.
Found in translation
Lingo.dev is the handiwork of CEO Max Prilutskiy and CPO Veronica Prilutskaya (pictured above) who announced that they sold a previous SaaS startup called Notionlytics to an undisclosed buyer last year. The duo had already been working on the foundations of Lingo.dev since 2023, with the first prototype developed as part of a hackathon at Cornell University. This led to their first paying customers, before going on to join Y Combinator (YC)’s fall program last year.
At its core, Lingo-dev is a Translation API that can either be called locally by developers through their CLI (command line interface), or through a direct integration with their CI/CD system via GitHub or GitLab. So in essence, development teams receive pull requests with automated translation updates whenever a standard code change is made.
At the heart of all this, as you might expect, is a large language model (LLM) — or several LLMs, to be exact, with Lingo.dev orchestrating the various input and outputs between them all. This mix-and-match approach, which combines models from Anthropic, OpenAI, among other providers, is designed to ensure that the best model is chosen for the task at hand.
“Different prompts work better in some models over other models,” Prilutskiy explained to TechCrunch. “Also depending on the use-case, we might want better latency, or latency might not matter all.”
Of course, it’s impossible to talk about LLMs without also talking about data privacy — one of the reasons that some businesses have been slower to adopt generative AI. But with Lingo.dev, the focus is substantively on localizing front-end interfaces, though it also caters to business content such as marketing sites, automated emails, and more — but it doesn’t funnel into any customers’ personal identifiable information (PII), for instance.
“We do not expect any personal data to be sent to us,” Prilutskiy said.
Through Lingo.dev, companies can build translation memories (a store of previously translated content) and upload their style guide to tailor the brand voice for different markets.
Businesses can also specify rules around how particular phrases should be handled and in what situations. Moreover, the engine can analyze the placement of specific text, making necessary adjustments along the way — for example, a word when translated from English into German might have double the number of characters, meaning that it would break the UI. Users can instruct the engine to circumvent that problem by rephrasing a piece of text so it matches the length of the original text.
Without the broader context of what an application actually is, it can be difficult to localize a small piece of standalone text, such as a label on an interface. Lingo.dev gets around this using a feature dubbed “context awareness,” whereby it analyzes the entire content of the localization file, including adjacent text or event system keys that translation files sometimes have. It’s all about understanding the “microcontext,” as Prilutskiy puts it.
And more is coming on this front in the future, too.
“We’re already working on a new feature that uses screenshots of the app’s UI, which Lingo.dev would use to extract even more contextual hints about the UI elements and their intent,” he said.
Going local
It’s still fairly early days for Lingo.dev in terms of its path to full localization. For example, colors and symbols may have different meanings between different cultures, something that Lingo.dev doesn’t directly cater to. Moreover, things like metric/imperial conversions is something that still needs to be addressed by the developer at the code level.
However, Lingo.dev does support the MessageFormat framework, which handles differences in pluralization and gender-specific phrasing between languages. The company also recently released an experimental beta feature specifically for idioms; for instance, “to kill two birds with one stone” has an equivalent in German that translates roughly into “to hit two flies with one swat.”
On top of that, Lingo.dev is also carrying out applied AI research to improve various facets of the automated localization process.
“One of the complex tasks we’re currently working on is preserving feminine/masculine versions of nouns and verbs when translating 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’s either “maestro” (male) or “maestra” (female). Making sure these nuances are preserved correctly falls under our applied AI research efforts.”
Ultimately, the game-plan is about much more than simple translation: It wants to get things as close as possible as to what you might get with a team of professional translators.
“Overall, the [goal] with Lingo.dev is to eliminate friction from localization so thoroughly, that it becomes an infrastructure layer and natural part of the tech stack,” Prilutskiy said. “Similar to how Stripe eliminated friction from online payments so effectively that it became a core developer toolkit for payments.”
While the founders most recently were based in Barcelona, they’re moving their formal home to San Francisco. The company counts just three employees total, with a founding engineer making up the trio — and this is a lean startup philosophy that they plan to follow.
“Folks at YC, myself and other founders, we’re all huge believers in that,” Prilutskiy said.
Their previous startup, which provided analytics for Notion, was entirely bootstrapped with high-profile customers including Square, Shopify, and Sequoia Capital — and it had a grand total of zero employees beyond Max and Veronica.
“We were two people, full time, but with some contractors for various things now and then,” Prilutskiy added. “But we know how to build things with minimal resources. Because the previous company was bootstrapped, so we had to find a way for that to work. And we are replicating the same lean style — but now with funding.”
