Translation Exchange — language as structured meaning

Translation Exchange began with the localization problem exposed by Geni’s Newsfeed. A story such as “Michael added David, Sarah and Rachel to the Smith Family Album” looked like a sentence, but underneath it were people, links, objects, gender, plurals, permissions, and language-specific structure.

Static key-and-string systems forced every language into the shape of the source sentence. TML took a different approach: describe the sentence’s meaning and structure, then let each language render it naturally.

That work became the tr8n framework and eventually the technical foundation of Translation Exchange.

While I was still developing tr8n at Geni, Alan Braverman—one of Geni’s early CTOs—walked past my desk, looked at the system, and made a comment that stayed with me.

At the time, tr8n was internal infrastructure. His reaction made the broader possibility tangible: perhaps sophisticated internationalization was not only a Geni problem.

The first major external validation came from Yammer. I integrated tr8n and used its community to crowdsource translations across more than twenty languages. At one point, Yammer effectively went fully Chinese overnight.

We also open-sourced the original tr8n framework. It attracted attention across the Ruby and Rails communities and created opportunities to test the model elsewhere.

Kongregate successfully integrated tr8n into its gaming platform. Geni could use it. Yammer could use it. Kongregate could use it. The technology was beginning to look like a platform.

The original framework remains available at github.com/berk/tr8n. I would build parts of it differently today, but it documents a new way of thinking about software localization at that moment in the web’s evolution.

Ian McDaniel, one of Geni’s top designers, and I co-founded Translation Exchange and launched the first version from Amplify.LA, which supported the idea early.

After gaining initial traction, we raised additional funding from Hillsven and Canyon Creek and expanded across engineering, infrastructure, sales, and marketing. My longtime friend Leon Kogan joined to build and operate our AWS infrastructure and CI/CD systems.

Then we started building—fast.

The original goal was to make Geni’s localization capabilities available to any software company. That meant moving far beyond a Ruby on Rails framework.

We developed SDKs and integrations across client, server, web, mobile, and native environments. Many of those libraries still live quietly under the Translation Exchange organization on GitHub.

The developer experience remained consistent: teams should build applications normally while Translation Exchange handled localization complexity around them.

Translation Exchange became much more than an internationalization library. SDKs discovered translatable content, synchronized it with the platform, and delivered approved language updates back into applications.

Localization became an ongoing operating system rather than a one-time development project. Developers, product teams, translators, reviewers, communities, and vendors could participate without sharing one code-centric workflow.

TML remained at the heart of the platform. A translation template described the meaning and structure of a sentence instead of treating it as an opaque string.

Because TML templates were more expressive than ordinary localization keys, conventional translation interfaces were not enough. We built a specialized environment that showed translators what tokens meant, which values affected grammar, what variations existed, and how the result would render inside the product.

Translators could work with the real meaning of dynamic software language without needing to understand the application’s source code.

Customers did not always want to recruit and manage translators themselves. Sometimes they simply wanted to ask the platform to translate an application into Japanese.

We connected Translation Exchange to professional providers including One Hour Translation, Gengo, and other agencies. SDK-collected content could become a translation request, quote, order, tracked project, and completed delivery without leaving the platform.

Translation Exchange became the layer connecting application developers, internal translators, communities, and professional language services.

Software companies repeatedly pay to translate the same concepts: Sign in. Save. Cancel. Welcome. Add a comment. Send message.

We asked why the world should translate the same sentence thousands of times. Customers could opt into a global translation space where compatible translations were shared across applications.

Instead of every product starting from zero, participating applications could benefit from localization knowledge that already existed across the network. The more applications participated, the more useful the memory became.

In many ways, we were trying to create a shared localization knowledge layer for software.

Through Hillsven and our investor Jean Kovacs, we were introduced to a team at LinkedIn. They had independently developed an internal localization framework with ideas that overlapped significantly with TML.

We held extensive technical discussions, product meetings, and integration conversations, and built prototypes demonstrating how Translation Exchange could fit LinkedIn’s infrastructure.

For a company our size, an adoption of that scale could have changed our trajectory. But large organizations reorganize, responsibilities move, and priorities shift. Internal changes and conflicts ultimately caused the opportunity to disappear.

Technically, we were extremely productive—perhaps too productive. We kept discovering interesting localization problems and solving them.

But startups are not rewarded for the number of interesting problems they solve. They are rewarded for finding a repeatable business.

Our biggest mistake was building too much too quickly while investing too little, too late, in customer acquisition and distribution. The product surface, team, infrastructure, and operating costs expanded faster than the customer base.

The LinkedIn opportunity amplified our strategic problem. Because the potential contract was so significant, we invested considerable time and energy in making it work.

We allowed one potential enterprise customer to become too important to the company’s strategy. When the opportunity disappeared, we did not have enough time or runway to replace it with a sufficiently large pipeline.

Once another funding round looked unlikely, we explored strategic alternatives with localization companies including Smartling and Gengo. None resulted in a transaction, and we eventually made the difficult decision to wind down.

Translation Exchange did not become the company we hoped it would, but the experience was enormously valuable.

Most importantly, it taught me that building something technically impressive and building a successful company are two very different problems. You need both.

I remain deeply grateful to the people who believed in Translation Exchange and gave us the opportunity to build it.

The company did not become what we envisioned, but the technology, relationships, and lessons had a lasting impact on everything I did afterward.

Although Translation Exchange shut down as a company, much of the technology remains publicly available. The original tr8n framework and Translation Exchange SDKs preserve several years of experimentation around a question that remains relevant.

Modern language models make that original question feel even more timely: how can software understand language as structured meaning instead of static text?

As Translation Exchange was winding down, Andrew Arrow—my friend and collaborator from Digital Evolution, Geni, and Yammer—showed me the Bird app before it launched and kept pushing me to take a closer look at the company.

That persistence led to Justin Balthrop—previously CTO at Geni and now CTO of a new company called Bird—inviting me to meet in Santa Monica and ride one of their scooters.

The scooters still had simple black-and-white logos that looked like wheels with wings. As we rode, people repeatedly stopped to ask where they could get one, whether they could rent it, and how much it cost. They were trying to pay us to ride scooters from a company they barely knew.

I remember thinking: They are onto something. Before Justin could really offer me a job, I essentially asked what I could do and when I could start.

That became the beginning of my Bird story.