An estimated billion people are currently studying a new language, yet most will never achieve conversational fluency. While applications like Duolingo have made language acquisition accessible and engaging, they rely primarily on recognition tasks—translating words, matching images, arranging tiles. The skill learners crave most—actual conversation—remains largely absent from these platforms.
Eevi, a voice-first AI language tutor, has secured pre-seed funding in a round led by Rockstart. The Amsterdam-based platform is constructed around spoken dialogue, employing real-time artificial intelligence rather than the swipe-and-tap mechanics prevalent across the industry. Co-founders Filemon Schöffer and Jago Gazendam established the company to address what they call the fluency illusion: learners accumulate impressive streaks and study habits, then freeze when confronted with actual conversation.
Schöffer, serving as CEO, previously co-founded Hubs.com, which Protolabs acquired in a deal valued at up to $330 million. He later held the position of Chief Commercial Officer at OpenUp, a European mental health platform, and co-authored The 3D Printing Handbook. Gazendam, the company's CTO, brings experience spanning education and healthtech sectors. He began his career integrating technology at NIST International School in Bangkok before launching multiple healthtech ventures. For Gazendam, Eevi represents a return to the education field.
The gap between recognition and production
Schöffer observes that "There are hundreds of millions of people trying to learn a second language, and only a small proportion will ultimately succeed. That's an enormous waste of human energy."
The industry's shortfall lies in the distance between recognising a language and actually producing it in speech. Until now, human tutors represented the primary alternative, though they come with higher costs and scheduling constraints. Eevi contends that real-time voice AI can bridge this gap.
From personal frustration to product vision
Teaching has long occupied Schöffer's thinking. "I've always been interested in teaching. My first company, when I was a teenager, was essentially an evening tutoring school for mathematics and physics." He found traditional schooling to be inflexible and repetitive.
Now that I have children myself, I see that again. I've always been interested in how we teach people and how people actually learn.
Filemon Schöffer
Following the sale of Hubs in 2021, Schöffer relocated to Japan to study Japanese. "I experienced firsthand how limited the existing apps were," he recounted. "I went to a language school, but I'm not particularly good at learning in group settings. I can't sit still! Again, I found myself thinking: somebody needs to fix this."
Gazendam, a childhood friend and technology entrepreneur, had moved to Spain and was similarly frustrated with the language-learning applications available. "We thought, with everything happening in AI, why don't we take a shot at language learning?" The pair commenced development in summer 2025 and introduced a closed beta in January 2026.
Beta insights reshaped the interface
Between January and June 2026, Eevi conducted a six-month closed beta involving approximately one hundred advanced learners. A critical discovery emerged around interface design.
Our original platform had the same basic chat interface as most LLMs. You could see a transcript of what you said and what Eevi said. But we discovered that reading while you're trying to speak isn't actually a very good experience.
Filemon Schöffer
This insight prompted a complete redesign centred on voice-first interaction. "Now you essentially have a blank canvas or whiteboard, and Eevi brings in supporting multimedia when it's useful. For example, it might say, 'We've practised these words, so now let's practise listening to them,' and bring an audio player onto the screen."
Rather than requiring learners to follow written transcripts continuously, the team developed multimedia components that Eevi can dynamically introduce to the whiteboard. Schöffer reflected: "We didn't know what a voice-first language-learning interface should look like when we started. Figuring out how it should work and feel was probably the biggest thing we learned during the beta."
How the platform operates
Schöffer notes that while "Duolingo has added some speaking, but its core intellectual property is the curriculum it has spent years developing. It can't simply abandon that model."
With Eevi, launching the application initiates a conversation with the tutor. Depending on proficiency, initial instruction may occur in the learner's native language or immediately in the target language. The platform aims to replicate the experience of working with a private language tutor, with the AI guiding the learner through each lesson.
It might say, 'This is your first Japanese lesson. Let's learn how to introduce yourself.' It can explain that there are different ways to introduce yourself in Japanese and provide the context for why, and then you actually have that conversation. From there, it's voice-first.
Filemon Schöffer
Eevi's curriculum integrates insights from more than 15 academic frameworks in language-learning research and encompasses more than 150 dynamic topics and situations, ranging from family interactions and grocery shopping to restaurant dining. Rather than prioritising grammar and memorisation, the team has emphasised rapid application of new knowledge. The curriculum concentrates on vocabulary and common expressions in realistic contexts.
According to Eevi, approximately 500 words, when mastered in context, account for roughly 80 per cent of everyday conversation.
Personalisation through continuous learning
Schöffer has long maintained that group-based learning is inherently one-size-fits-all and that this model has become obsolete. "I have this slightly cartoonish view that everybody will have a personal tutor in their pocket. Language learning is commercially attractive, so I think it's one of the first skills where we'll see this happen. Whenever we build something new for the Eevi curriculum, I ask myself: if this were physics, would this approach still work?"
As learners progress, Eevi simultaneously learns about them. Each conversation deepens the system's understanding of individual challenges and interests, informing subsequent lessons.
Currently, during onboarding, learners indicate their level using frameworks such as the European CEFR system or by identifying themselves as beginners. As lessons progress, Eevi tracks taught material and learner responses. The company has constructed an individual memory profile for each user, which Schöffer views as a distinct advantage over generic chatbot practice.
The platform intends to advance this further with conversational assessment. Rather than self-reporting proficiency, initial interaction would involve the AI asking questions in the target language, listening to responses, and dynamically evaluating ability. Schöffer recognises AI's capacity to customise learning at scale as a fundamental advantage over static curricula.
If you tell Eevi, 'This is going too slowly,' it can say, 'Okay, let's skip the next two modules and see how you go.' If that's suddenly too difficult, you can take a step back. You can guide your own tutor.
Filemon Schöffer
I think that's fundamentally different from a static curriculum. That's why we have these two elements working together: the dynamic curriculum and the conversation.
Filemon Schöffer
Constraints of current AI models
Eevi presently relies on Gemini as its voice model across approximately 14 languages, including Japanese, Indonesian, and Chinese. A recent Gemini update will expand this to around 35 languages. However, dependence on foundation models constrains which languages and dialects the platform can support.
When asked about handling accents, dialects, and pronunciation, Schöffer acknowledged: "Different LLMs support different dialects, and that's actually raised a much bigger issue for us." He perceives an element of neo-colonialism in LLM development: predominately Western-created models effectively determine which dialects receive representation and support.
"We've already had users saying, 'My mother speaks this dialect, and I want to learn it,' but the model doesn't support it." The company is investigating whether Eevi's underlying technology could also enable support for dialects and endangered languages, potentially as a separate product built on the same foundation.
Generative AI also introduces vocabulary and grammar errors, which Schöffer acknowledges currently prevents Eevi from offering certain niche languages. An LLM might technically support a language without possessing sufficient contextual understanding, raising error likelihood. However, he added: "We don't want to remain permanently dependent on the major LLM providers, particularly as a European company."
The long-term strategy involves developing its own model with greater sophistication around niche and protected languages, using that knowledge to teach them. "But first we need to demonstrate commercial success in mainstream language learning."
Lessons from building in the AI era
As a repeat founder, Schöffer observes that building during the AI era feels fundamentally different. "Normally, I'd say I know who to hire and how to approach certain problems, but I'm not sure I do anymore." Yet experience has clarified what merits immediate attention and what can wait.
"When we started Hubs, I worked 60-plus-hour weeks, year after year. Looking back, half of those hours were probably spent on things that weren't particularly important. When you don't know what to do, sometimes you just put your head down and work harder." With young children now, he is approaching Eevi with different priorities.
I know the product needs to be right, and we're deliberately not doing everything as quickly as possible. In the past, I would have launched fast, pushed fast, and started advertising fast.
Filemon Schöffer
Initial market focus
Beta testing validated Eevi's initial target audience: migrants and international professionals. "We're a B2C app, so anyone can use Eevi, but initially we want to start small and focus on international professionals who want to progress beyond something like Duolingo but don't want to attend a traditional language school because of the price or lack of flexibility. We're positioning ourselves somewhere in the middle."
Pricing includes a six-month course at €49 and monthly access at €18. The company notes that six months of Eevi costs approximately the same as a single hour with a human tutor, depending on the tutor's rates. Corporate B2B sales are also planned.
"I know the Amsterdam startup and scaleup ecosystem very well, so we'll approach companies and ask whether they'd be interested in buying, say, 50 seats for international employees who want to learn Dutch. That's something we'll start exploring."
Competitive positioning
Schöffer categorises the competitive landscape into three segments: established language-learning applications, foundation-model providers, and AI-native language-learning startups.
Regarding established players such as Duolingo, he stated: "I'm not too concerned about them because they're so heavily invested in their existing models. That's the classic innovator's dilemma. Of course, you never know, but I don't necessarily see them taking the lead in this approach."
At the other end sit LLM providers themselves. Could Gemini or OpenAI develop language-learning features directly atop their models? "I'd be surprised," he said. "They're effectively in the token business, and I don't necessarily see why they would build every vertical application themselves."
Between these extremes are companies like Eevi creating dedicated AI-native language-learning products. "I've seen more of those in the US than in Europe so far," Schöffer shared.
The fundraising decision
The founders bootstrapped Eevi for nearly 12 months and, according to Schöffer, possessed the capacity to continue self-funding. Yet in a rapidly evolving market, he believes velocity outweighs equity dilution. "And the right investors bring things personal capital does not: they pressure-test your thinking, open their network for the hires ahead, and hold you accountable. Rockstart backed us at 3D Hubs before anyone else did. Having them do it again is worth more than just the check."
Max ter Horst, Managing Partner of Rockstart, noted that as AI transforms education and employment, conversational foreign-language abilities will grow increasingly vital to economic mobility, inclusion, and international cooperation. "Eevi is building a new kind of language learning experience centred around conversation, confidence, and accessibility. We're thrilled to back this exceptional team, which includes a founder with whom we've successfully partnered before."
Schöffer acknowledges that prior success provides fundraising advantages. "I'm in a fortunate position because I already know a lot of VCs and they'll respond to me. But with the way AI is developing, I've found that a lot of VCs are hesitant about two things: software in general and B2C. If you say you're building a B2C language-learning product, the door can close pretty quickly. VCs tend to work from certain checklists, and I've heard 'come back when you have €1 million in revenue' quite a lot. My reaction is: if I've already reached €1 million in revenue, I hope we're doing pretty well!"
The funding will support Eevi's public launch and expansion throughout the coming year. Schöffer aims for the company to demonstrate that AI can enable a fundamentally transformed approach to learning. "I want people to recognise that learning — and in our case, language learning — can be significantly better. There are hundreds of millions of people trying to learn a second language, and only a small proportion will ultimately succeed. That's an enormous waste of human energy. The platform is new, and it's still evolving, but I hope people use it and see the potential for a different way of learning."
Source: Tech.eu
