Panels: Audio Data for AI Labs

Hi!

We’re Aaron and Jason, and we’re building Panels.

TL;DR — Panels builds diverse, high-quality conversational audio datasets for training Voice AI models. If you or someone you know is working on voice and needs audio data, we’d love to talk!

https://www.youtube.com/watch?v=auANpScmgls

Our Story

Aaron and Jason met at Princeton University where they both studied Operations Research and Financial Engineering.

• Aaron is a 1x world champion and 5x national champion rower. After leaving rowing, Aaron became a 2x national champion in cycling, racing on a professional team against former Tour de France winners.
• Jason previously helped build and grow startups to millions of users and worked on the data team at Amazon. He also taught NLP at Princeton.

The problem

Voice AI is exploding: in the past three YC batches, 60+ companies have depended on voice models. But voice models live or die on the quality of their training data.

Labs need datasets that reflect:

• Languages (global coverage, not just English)
• Accents (real-world variety)
• Emotions & expressiveness (not monotone samples)
• Demographics (ages, genders, cultural nuance)

Right now, most labs struggle to source enough realistic, diverse, high-quality conversational audio.

Our solution

Panels delivers exactly that.

• We have a network of 10,000+ vetted voice contributors
• Coverage across 20+ languages and 100+ countries
• Expressive, conversational audio datasets designed for training state-of-the-art voice models.

This makes Panels the reliable data provider for the labs building the future of Voice AI.

Who this is for

• Voice labs scaling foundational models
• Research institutions working on speech, linguistics, and AI
• Conversational AI startups wanting to finetune their agents

Our ask

If you (or someone you know) is building in voice and needs high-quality audio data, we’d love to talk.

Email us at 📩founders@usepanels.com
 or ☎️Book a call