Pavel Ahafonau spent five years trying to learn Lithuanian. He tried apps, courses, and teachers. Nothing worked. Then in early 2020, with a government language exam looming and a newborn at home, he did what engineers do when they can’t find the right tool: he built one.
Five months later, he passed the exam on his first try.
That tool became WRD – a mobile-first, AI-powered language learning accelerator that, according to the company, has crossed two million users, surpassed $1.2 million in annual recurring revenue, and grown 3.5x in six months. The company is now seeking a seed round of $3 to $5 million. But the story behind WRD stretches back decades, through IBM supply chains.
The engineer who couldn’t stop building
Ahafonau has been writing code since 1986. He started his professional career at IBM in Belarus in 1998, before relocating to Germany in 1999, where he spent several years working on large-scale enterprise systems. He says this included work on a global supply chain and ERP integration platform connecting IBM manufacturing plants across multiple continents. The work was complex, high-stakes, and deeply technical. It was also, in a sense, the beginning of a pattern: find a broken system, understand it at a fundamental level, and rebuild it from scratch.
By 2010, Ahafonau had co-founded Happymagenta, a mobile studio he started after noticing that his personal apps were generating serious revenue. The company says he later recruited a former IBM colleague with a simple pitch: the numbers are working. He said he wanted to build something bigger. What followed was a decade of mobile products – network diagnostic tools, navigation apps, word games – that reportedly built large audiences over time. Spyglass is the studio’s AR navigation app inspired by fighter jet heads-up displays. Tomb of the Mask is a fast-twitch arcade game that has shown notable success.
What Ahafonau built alongside these products was less visible but arguably more valuable: a deep operational knowledge of what makes mobile software actually work. The company says it built a custom CDN infrastructure spanning 15 global data centers, real-time funnel optimization system that adjusts app interfaces across billions of possible configurations automatically. It developed an App Store keyword optimization algorithm, based on a variant of the Traveling Salesman Problem, which increased organic search traffic by more than 30%. These weren’t products. They were competitive infrastructure, built quietly and kept private.
“Everything I know from those years went into WRD,” Ahafonau says.
The problem with language apps
When Ahafonau set out to learn Lithuanian in 2020, he installed every major app on the market. None of them worked for him. The core problem, as he saw it, wasn’t content – it was structure. Many existing apps have traditionally relied on fixed or structured curricula, moving them through predetermined lessons at a predetermined pace, regardless of what they already knew or how quickly they were progressing.
“You open it, and it starts: hello, welcome. It doesn’t matter what level you are,” he says. “It wastes your time.”
What Ahafonau had at his disposal was something most language app developers didn’t: years of research into how words actually function in real human communication. That research had been conducted originally to power the word games Happymagenta was building — understanding word frequency, idiom density, and how language is used in books, films, encyclopedias, and everyday conversation, rather than how it appears in textbooks. The insight was that a relatively small set of high-frequency idioms may help unlock a disproportionately large share of comprehension. Some language-learning advocates argue that learning the top 1,000 idioms may help learners understand a significant share of everyday content — sometimes estimated at 60 to 80 percent, depending on the language.
From that foundation, the team built WRD’s core algorithm: a personalized learning curve that detects a user’s existing knowledge within the first 30 to 100 words, places them precisely at their actual level, and then continuously recalculates the optimal next step based on their performance. No fixed lessons. No unnecessary repetition. No starting from scratch.
WRD’s own research, published on the app’s website, shows that users following this approach reach the same levels of real-world language comprehension up to 2.8 times faster than with conventional apps — saving, in some cases, more than 160 days of study time.
Flow state as a product feature
Speed of learning is only part of what Ahafonau is engineering. The other part is harder to articulate but central to how WRD works: keeping users in a state of effortless engagement.
The team draws heavily on behavioral techniques refined across years of game development. A minimum learning session in WRD takes 2.5 minutes. Users can operate the entire app one-handed. A swipe gesture replaces character-by-character typing. Easy words aren’t hidden — they’re shown intentionally, giving users an emotional reward and keeping the algorithm calibrated. The difficulty curve is tuned to feel neither frustrating nor boring.
“Our goal is that the user stops thinking about what they’re doing,” Ahafonau explains. “When they stop thinking, that’s when real learning begins.”
He draws an analogy to driving: the most fluent drivers aren’t consciously executing every movement. They’ve internalized the skill. WRD is designed to accelerate that same process for language – using engagement mechanics from gaming to lower cognitive resistance and let the algorithm do its work below the level of conscious effort.
The approach produces unusual retention numbers. According to the company, 10–12% of app users continue paying a year later, while returns account for less than 1% of annual transactions. “We don’t try to take as much money as possible from users as fast as possible,” Ahafonau says. “We make more money from users who stay with us longer.”
A small team, a large machine
WRD is built by four developers, including Ahafonau himself, who remains hands-on with both algorithm design and infrastructure. He is currently working on two things simultaneously: retraining the learning path models using accumulated user data, and integrating the same real-time funnel optimization system — what the team calls Magic – that Happymagenta has used in its games for years.
The company says that in games where the system has been deployed, user retention at level 40 has exceeded retention levels some competing games see at level 10. The system works by running millions of interface variants simultaneously and automatically converging on the highest-performing configuration – not through sequential A/B tests, but through continuous, in-flight optimization across the entire user funnel at once. Bringing that into WRD is, Ahafonau believes, the next significant growth lever.
The company says it reached $1 million ARR in July 2025 and crossed $1.2 million by December, representing 200% growth over the preceding six months. From late summer 2025 through early 2026, according to the company, the app grew 20 to 30 percent month over month, with revenue increasing fourfold over six months. It is now preparing a new content expansion to serve more advanced learners, and will soon add phrase-building and conversational features – moving WRD beyond vocabulary acceleration into the full arc of language acquisition.
The company is headquartered in Lithuania and is currently raising a seed round of $3 to $5 million. Ahafonau is direct about the dynamic with investors: by the time most finish their diligence process, the terms they originally offered no longer reflect where the company is.
He says the company had $300K ARR when discussions began and had reached $1.3M by the time talks resumed, which he described as a fast-moving opportunity. “By the time they came back, it was $1.3M. The window closes.”
For a founder who once passed a government language exam by building the tool he needed himself, that kind of momentum feels less like a risk and more like a feature.
The information provided in this article is for general informational and educational purposes only. It is not intended as financial advice. Readers should not rely solely on the content of this article and are encouraged to seek professional advice tailored to their specific circumstances. We disclaim any liability for any loss or damage arising directly or indirectly from the use of, or reliance on, the information presented.
VentureBeat newsroom and editorial staff were not involved in the creation of this content.