Announcing Our $3.8M Seed Round to Make Voice AI More Reliable
We raised $3.8 million. Mischief led the round. YCombinator, AI Grant, Pioneer, Coalition Operators, Coughdrop, NEA, and a bunch of other great funds participated—plus 30+ angels who've built and sold AI companies themselves.
I'm not going to pretend this was inevitable. When we started, voice AI testing wasn't even a category. We'd pitch investors and spend half the meeting explaining why it mattered. The ones who got it immediately were people who'd actually tried to ship voice agents into production and discovered how painful it is when things break and you can't figure out why.
Quick filter: If you've ever watched a "perfect" demo break in production, you already understand why we're building this.
Why Voice AI Testing Matters Now
Here's what we kept hearing from teams building voice agents: "We're spending more time testing than building." Engineers would manually dial their own agents dozens of times, trying different accents, background noise levels, edge-case questions. Then they'd ship, and within a week, customers would find bugs they'd never thought to test for.
The math doesn't work. You can't manually test every scenario. Change your STT provider? Your carefully-tested agent might start mishearing numbers. Update a prompt? Something that used to work stops working. It's whack-a-mole, except the moles are customer-facing and the stakes are real.
And after deployment? Operations teams at some of our customers were listening to hundreds of calls daily trying to catch problems. That's not sustainable. It's expensive, it's slow, and issues slip through anyway.
What We're Building
We built AI voice agents that test other AI voice agents. Sounds recursive, but it works. Our test agents can simulate thousands of conversations simultaneously—different accents, background noise, interruptions, edge cases. The kind of testing that would take a human team weeks happens in minutes.
Beyond testing, we're doing production monitoring. When something goes wrong on a real call, we flag it. When patterns emerge—like a specific phrase causing confusion 15% of the time—we surface that.
The goal: you should know about problems before your customers complain about them.
What's Next
If 2024 was the year of the prototype, 2025 is going to be about reliability. The companies shipping voice agents into healthcare, finance, insurance—they're operating under compliance frameworks that weren't written with AI in mind. That's changing. And when it does, "we tested it manually a few times" won't cut it anymore.
Lauren Farleigh from Mischief said it better than I could: testing and governance tools haven't caught up to what developers actually need. That's the gap we're filling.
A Bit About Me
Before Hamming, I ran trust and safety infrastructure at Citizen (the Founders Fund-backed personal safety app). Before that, I was at Tesla building out their AI-powered sales program. Both jobs taught me the same lesson: when systems are customer-facing and high-stakes, reliability isn't optional. You either build the infrastructure to catch problems early, or you deal with the consequences when they hit production.
That's why we built Hamming. Voice AI is going to be everywhere. The testing infrastructure has to be there first.
If you're building voice agents and want to talk about testing, or if you're interested in joining the team, email me directly at sales@hamming.ai. Always happy to talk shop with people who've felt this pain themselves.
— Sumanyu
