Panels

Panels is purpose-built for voice AI teams that have moved past prototyping and need data that actually fixes model gaps. The consultative process — research, collection, iteration — makes sense: you're not buying a CSV dump, you're commissioning data that targets your specific use case. The proprietary multilingual speaker-separated dataset is a real moat; getting that right at scale is hard. For a startup building a voice assistant for a niche accent or a lab evaluating turn-taking models, this could be exactly what's missing. But the lack of public pricing is a friction point. Small teams or bootstrapped founders won't know if they can afford it without a sales call. And if you just need a quick benchmark or off-the-shelf corpus, you're better off with public datasets or self-service platforms like Mozilla Common Voice or Kaggle. Compared to competitors like Appen or Defined.ai, Panels is more focused (audio only) and more hands-on. We'd reach for this when data quality directly impacts model performance — think medical dictation, multilingual assistants, or safety-critical voice interfaces. Just budget for the time investment: the iterative phase means you'll be in a feedback loop for months, which is a feature if you're iterating on a model, but a distraction if you just want to check a box.