The folks at Vatis Tech had built something remarkable: AI transcription that actually understood Romanian, Bulgarian, and other underserved languages with accuracy matching human transcribers. But their interface looked like it was built by developers for developers, because it was.

The complexity was overwhelming new users while enterprise clients struggled to access the advanced features they needed. The website was converting at just 1% despite superior technology to competitors. They were losing to consumer-friendly tools like Otter.ai and Rev that had inferior accuracy but superior UX.

...and the ultimate constraint: zero downtime allowed. We had to redesign while flying the plane.

As a Lead Designer, I've owned the complete product and website redesign from ground up for the past 2 years. This includes the website architecture and implementation in Webflow, transcription platform redesign, and stakeholder management across technical and business teams.

The unique challenge was balancing the needs of data scientists who built the AI with business users who just wanted fast and accurate transcripts.

I started with forensic analysis of where users struggled. The analytics revealed shocking drop-offs: users dropped off after file uploads, most never returned after first transcription, and average time to first successful transcription was measured in minutes, not seconds.

The interface exposed every possible parameter the AI could handle, overwhelming users who just needed accurate text. Enterprise users complained it looked like it was built by developers for developers. Which it was.

I analyzed the entire competitive landscape to understand what users expected. Rev kept it dead simple but offered limited features and no data privacy. Otter.ai had great UX but weak support for non-English languages. Descript was powerful but overwhelming, focused more on editing than transcription. Assembly AI was purely developer-focused with no UI for business users.

‍Key Insight — The average user doesn't care about AI technology, they care about getting accurate transcripts quickly in their language. If you can integrate AI without it feeling overwhelming, you've got a winning formula.

Instead of hiding complexity or dumbing down the product, we worked on developing ways to reveal it progressively based on user maturity.

The solution was the Workspace Concept: Personal workspace with just 3 settings and smart defaults alongside sample files, API Dashboard with full API access, visual configuration, and advanced testing tools, as well as an Enterprise workspace with custom workflows, SSO, and compliance tools.

The biggest challenge was redesigning while keeping enterprise clients happy. We couldn't just flip a switch.

We've developed a dual-track approach: Track 1 maintained the current platform and fixed critical bugs, Track 2 built the new experience behind feature flags, with gradual migration supported by the customer success team and a one-click fallback to the classic interface for anyone who needed it.

The transformation was dramatic. First upload moved from behind a login to letting users try without an account, directly on the website.

Language selection became automatic through AI detection with easy override. Time to first transcript dropped from 12 minutes to under 60 seconds.

Smart defaults now learn from usage. The AI detects language from the first 10 seconds of audio. Previous settings are remembered per workspace. The system suggests optimal settings based on audio quality and shows confidence levels for transparency so users understand why certain settings are recommended.

Creating cohesive experience across all touchpoints was crucial for building trust. The marketing website built in Webflow sets the expectations. The web application in React delivers on those promises. Developer API documentation speaks the same visual language.

Using Webflow, I architected and implemented Vatis Tech's entire web presence to support their global ambitions. The challenge was supporting 40+ language markets while maintaining local relevance.

Built together with the Vatis Tech Marketing Team, the website includes 10,000+ pages alongside a sophisticated CMS architecture—programmatic SEO creates landing pages for language combinations like "Translate Polish to Romanian", industry verticals from legal to education, and specific use cases from subtitles to meeting notes. Trust signals include security badges, testimonials, and case studies relevant to each market. All combined with Webflow-native multi-language implementation.

The performance metrics speak for themselves: 92+ PageSpeed scores globally despite rich content, expandable organic reach to 40+ language markets, 3x increase in organic traffic in the past year alone, page 1 rankings for hundreds of programmatic SEO pages, and 65% of new accounts now coming from organic search versus almost 0% just two years ago.

The three-click promise became our north star. No matter how complex your needs, you're three clicks from results: Drop file or start recording, confirm settings or use smart defaults, get transcript.

For power users and developers who need advanced features, API Playground is available. With custom AI instructions, speaker identification, confidence thresholds, formatting rules, webhook integrations, batch processing, and more. The key was making these features discoverable without overwhelming new users.

This project reinforced my belief that great design makes complex technology feel simple. The challenge wasn't simplifying Vatis Tech's AI. It was revealing its power progressively, letting users discover advanced features as they grow.

‍Key Learning — Progressive disclosure beats feature hiding. Show users what they need when they need it. Smart defaults eliminate 80% of user decisions, making complex tools accessible to everyone. Showing confidence levels, not hiding complexity, builds trust with users. Localization is more than translation—it's about understanding local needs and expectations.

Finally, the website is often the first product experience and should be designed with the same care as the app itself.