ConvoPro AI — Designing Psychological Safety at Scale

Users described it as “motivating,” “realistic,” and “helpful without pressure.”

Babbel had powerful tools for grammar and reading. But when it came to the thing learners wanted most — actually speaking — engagement collapsed. Especially in high-stakes moments like live classes, job interviews, or certification exams.

Working closely with product, data, and learning science leads, I helped identify a critical insight that would reshape our approach: learners knew the grammar. They understood the vocabulary. But the moment they had to speak — they froze. This wasn't a content gap. It was an emotional one.

And the business case was clear: if we reduce friction in speaking practice, we can boost activation and retention — without adding tutor costs.

This wasn't a project handed to me with a brief. I shaped it from the ground up. I led the design vision and drove the product direction in close collaboration with PM, engineering, and didactics. My work spanned three interconnected layers:

This tension — high desire, low engagement — became our strategic starting point. It told us the problem wasn't motivation. It was something deeper.

Babbel's existing speaking features relied on repetitive, scripted exercises. They taught pronunciation — but not fluency. Not the ability to think on your feet in a real conversation. Advanced learners were stuck in a loop: they could understand everything, but they couldn't respond without panic.

“I know what to say. But when I try to speak, I just freeze.”
“It's not about grammar — it's about pressure.”

There's a name for this in linguistics: the silent period — the cognitive bottleneck where understanding doesn't translate into fluency. And existing tools weren't helping:

To move from assumptions to real understanding, I co-led in-person, moderated research sessions with 6 advanced learners in Berlin. We focused on four questions:

Before committing to a full product direction, I pushed for a lean validation step. We ran quick, targeted surveys with existing Babbel users to pressure-test our qualitative findings at a broader scale. The goal wasn't deep research — it was fast signal.

The fastest path to good decisions is often the smallest possible test, run at the right moment.

Through cross-functional planning sessions, I led the synthesis of our research into measurable product objectives — bridging learner needs, product strategy, and business goals.

Rather than jumping to a single solution, I facilitated a structured brainstorming process with PM, engineering, content design, and learning science. We explored three distinct concepts — each rooted in our research insights but taking a fundamentally different approach.

This is how I approach product design: diverge intentionally, then converge with evidence.

With principles defined, I moved into rapid prototyping — building mid- and high-fidelity flows in Figma and Protopie to test different interaction models and feedback patterns.

In early prototypes, we tested showing corrections during the conversation. The result? Learners became more self-conscious and stopped speaking naturally. We shifted to a post-conversation summary model — designed to feel like coaching, not grading.

5 out of 6 test participants preferred the summary model. It felt like encouragement, not evaluation.

Many users hesitated at the initial “tap-to-start” mic input. They weren't sure when to speak, or what would happen. I redesigned the interaction to mimic a familiar mental model — WhatsApp's tap-and-hold voice input. We added micro-animation on first use and a contextual tooltip.

A beautiful interface means nothing if the AI behind it sounds robotic. To make ConvoPro sustainable and scalable, we needed a system — not just a feature. I co-designed a collaborative QA workflow that aligned learning science, content design, and engineering around a shared definition of quality.

This feedback loop didn't just improve the product. It established a new way of working — treating prompt tone, pacing, and phrasing as UX artifacts with the same rigor as interface design.

The final product was more than a chat feature. It was designed to help learners practice speaking in a way that felt safe, structured, and motivating.