Our team has been part of the German AI scene since the 2010s. Over the past decade, we've watched the industry evolve from clunky experimental chatbots to the sophisticated, multi-modal Voice AI agents now running critical business operations.
But through all those years of building production-level systems, one problem kept coming back to haunt us: what we call the Evaluation Gap.
Getting a voice agent to shine in a controlled demo? That's a solved problem. Getting it to stay safe, consistent, and compliant across thousands of unpredictable real-world customer calls? That's a completely different beast. For too long, figuring out whether an agent is actually ready for production has been a manual, fragmented, and frankly unreliable process.
Today, we're launching Wir_Schwatzen in Nürnberg to change that.
The Challenge: Flying Blind with AI
Voice AI is fast becoming the primary interface for businesses worldwide, yet many organizations are still essentially guessing how their agents will behave when things get messy. That guesswork creates three real risks:
1. Regulatory Exposure
The EU AI Act is now in effect, and where and how you process data is no longer optional; it's a legal requirement. Routing sensitive voice data and proprietary information outside Europe has quietly become a serious compliance problem.
2. The QA Bottleneck
Manual testing simply can't keep up with how fast AI gets deployed. Without automation, teams are stuck choosing between slow launches and shipping agents nobody's properly vetted.
3. Lost in Translation
Most evaluation tools were built on non-European datasets. They routinely miss the linguistic quirks and cultural expectations that matter most in our regional markets.
Our Mission: Taking the Guesswork Out
At Wir_Schwatzen, we build the automated infrastructure that lets you move a voice agent from the lab to a live customer line and actually trust it. We focus on three things:
- Data Stays in the EU: Test calls are processed and stored in the EU. Personal data in transcripts and audio is redacted by default, and you decide how long data is kept.
- Repeatable Tests, Not One-Off Calls: We swap manual test calls for scenarios that run the same way, several times, on every release. Your team gets to spend its time building, not dialing.
- A Clear Verdict for Everyone: Product Managers and CX teams get a plain-language verdict, engineers get the evidence behind it, down to the exact moment in the call. No engineering bottleneck required.
How It Works
The Wir_Schwatzen platform is built as a decoupled system designed for serious benchmarking:
- Synthetic callers phone your agent at its number, using different personas with varying speed, interruptions and background noise, and throw edge cases at your system to find exactly where it breaks.
- The platform analyzes every call: transcript, measured latency and silences per turn, and a verdict for each scenario with its reasoning and evidence.
Built in Nürnberg, Built for Europe
The name Wir_Schwatzen captures what we're really after: making conversations with AI feel as natural and dependable as talking to a person.
After ten years in the trenches of AI development, we're turning our focus to the part that matters most, knowing the truth about how your agent actually performs. It's time to stop guessing and start measuring.