Lingo.dev - App Localization Engine for Developers

The Expanding Accessibility of Global Communication

Individuals focused on a single language now possess unprecedented ease in reaching a worldwide audience. Google Translate, a long-standing resource, is capable of converting content from images, audio recordings, and complete websites into numerous languages.

Furthermore, emerging tools like ChatGPT function effectively as portable translation assistants.

The Rise of AI-Powered Language Technologies

Companies like DeepL and ElevenLabs have achieved substantial valuations, exceeding a billion dollars, due to their advanced language-based technologies. These innovations are being integrated into a variety of business applications.

However, a new competitor is emerging, offering an AI-driven localization engine designed to provide the necessary infrastructure for developers aiming for global reach.

Introducing Lingo.dev: Localization for Developers

Previously operating as Replexica, Lingo.dev focuses on developers who desire complete front-end localization for their applications.

Developers can continue their standard coding practices, while Lingo.dev operates automatically in the background.

This eliminates the need for manual text transfer using tools like ChatGPT, or the complexities of managing multiple translation files from various agencies.

Current Clients and Funding

Lingo.dev currently serves notable clients including Mistral AI, a French unicorn company, and Cal.com, an open-source alternative to Calendly.

To facilitate further expansion, the company has secured $4.2 million in seed funding.

This funding round was led by Initialized Capital, with additional participation from Y Combinator and a group of angel investors.

Key benefits of Lingo.dev include:

• Automated localization process
• Elimination of manual translation tasks
• Streamlined workflow for developers
• Integration with existing development pipelines

A New Approach to Translation

Lingo.dev was created by CEO Max Prilutskiy and CPO Veronica Prilutskaya (shown above). They revealed that their prior SaaS company, Notionlytics, was sold to an unnamed purchaser the previous year. Development of Lingo.dev’s core components began in 2023, initially as a hackathon project at Cornell University. This initial work quickly attracted their first customers, leading to acceptance into Y Combinator’s fall program last year.

Essentially, Lingo.dev provides a Translation API. Developers can utilize it either locally via their CLI (command line interface) or integrate it directly into their CI/CD systems through platforms like GitHub or GitLab. This means development teams receive automated translation updates with each standard code modification submitted as a pull request.

The system’s functionality relies on a large language model (LLM) – or, more accurately, multiple LLMs. Lingo.dev manages the interactions and outputs between these various models. This flexible approach, incorporating models from providers such as Anthropic and OpenAI, aims to select the optimal model for each specific task.

Prilutskiy explained to TechCrunch that “certain prompts perform more effectively with some models than others.” He also noted that “depending on the application, either lower latency is desired, or latency may not be a critical factor.”

Data privacy is a crucial consideration when discussing LLMs, and a reason for some businesses’ cautious adoption of generative AI. However, Lingo.dev primarily focuses on localizing front-end interfaces. It also supports business content like marketing websites and automated emails, but it is designed to avoid processing customers’ personally identifiable information (PII).

“We do not anticipate receiving any personal data from our users,” Prilutskiy stated.

Companies using Lingo.dev can establish translation memories – repositories of previously translated content – and upload their style guides to maintain consistent brand messaging across different markets.

Businesses can also define specific rules for handling particular phrases and their usage. Furthermore, the engine analyzes the placement of text, making adjustments as needed. For example, a German translation of an English word might require more characters, potentially disrupting the user interface. The engine can be instructed to rephrase the text to match the original length.

Localizing isolated text, such as a user interface label, can be challenging without understanding the application’s overall context. Lingo.dev addresses this with a feature called “context awareness.” It analyzes the entire localization file, including surrounding text and event system keys often found in translation files, to grasp the “microcontext,” as Prilutskiy described it.

Further enhancements to this capability are already in development.

“We are currently working on a new feature that utilizes screenshots of the application’s UI. Lingo.dev will then extract additional contextual clues about the UI elements and their intended function,” he revealed.

Localizing with Lingo.dev

Lingo.dev is still developing its comprehensive localization capabilities. Currently, the platform doesn't directly address cultural variations in the interpretation of colors or symbols. Furthermore, the implementation of metric and imperial unit conversions remains a task for developers to handle within the codebase.

However, Lingo.dev incorporates support for the MessageFormat framework. This framework effectively manages variations in pluralization rules and gendered language across different languages.

Recently, an experimental beta feature focused on idioms was also launched. For example, the English idiom “to kill two birds with one stone” has a German equivalent that translates to “to hit two flies with one swat.”

Ongoing Research and Development

Lingo.dev is actively engaged in applied AI research to enhance various aspects of the automated localization process. This research aims to improve the accuracy and nuance of translations.

“A significant challenge we are addressing is the preservation of gendered forms of nouns and verbs during translation,” explained Prilutskiy. “Languages differ in the amount of information they encode. The English word ‘teacher’ is gender-neutral, while its Spanish counterparts are “maestro” (male) or “maestra” (female). Ensuring these subtleties are maintained is a key focus of our AI research.”

The ultimate objective extends beyond simple translation. Lingo.dev strives to achieve results comparable to those produced by a team of professional human translators.

A Vision for Seamless Localization

“Our overarching goal is to eliminate the complexities of localization, integrating it as a fundamental infrastructure layer within the technology stack,” Prilutskiy stated. “We aim to achieve a level of seamlessness similar to that of Stripe, which revolutionized online payments and became a core tool for developers.”

Although the founders were recently located in Barcelona, they are establishing their official headquarters in San Francisco.

The company currently operates with a small team of three employees, including a founding engineer. This reflects a deliberate commitment to a lean startup approach.

“This philosophy is strongly supported by the Y Combinator community and many founders,” Prilutskiy noted.

Bootstrapped Success and Future Growth

Their prior venture, an analytics platform for Notion, was entirely self-funded. It attracted notable clients such as Square, Shopify, and Sequoia Capital, while being managed by only Max and Veronica.

“We operated as a team of two, full-time, occasionally utilizing contractors for specific tasks,” Prilutskiy clarified. “We have proven our ability to build effectively with limited resources. Having successfully bootstrapped our previous company, we’ve developed strategies for efficient operation, and we are applying the same lean principles – now with the benefit of funding.”

• MessageFormat: A framework used for handling pluralization and gender-specific phrasing.
• Applied AI Research: Ongoing efforts to improve the accuracy and nuance of automated localization.