Future AGI

Hey PH community! Nikhil here, Founder & CEO at Future AGI. Today we're launching SIMULATE, and I'm genuinely excited to share what we've built.

Why we started this?

Working with AI teams over the past few years, I kept seeing the same problem everywhere: voice AI testing was completely manual. Teams would build these amazing voice agents, then spend weeks manually calling them over and over, trying to find where they break.

It hit me when I watched a team spend a month testing their voice agent, only to have real users find critical bugs within the first day of launch. We knew there had to be a smarter way.

How it actually works?

SIMULATE creates AI agents that automatically call your voice AI system and have real conversations with it. But here's the key difference - instead of just checking transcripts like other tools, we built our evaluation models to analyze the actual audio.

Why does this matter? Because voice AI doesn't just fail in the words it says. It fails in tone, audio quality, response timing - stuff you completely miss if you're only looking at text transcripts.

The technical bits

We can spin up thousands of these test conversations simultaneously. Each one has different characteristics - different accents, background noise, conversation styles. It's like having a massive team of testers, but they never get tired and they explore scenarios no human would think of.

Setup is dead simple. Just give us your voice agent's phone number and we start testing immediately. No complex integrations needed - works with multiple integrations like Vapi, Retell or whatever platform you're using.

Where we're headed?

We're already working on multiple modalities and much more. The bigger vision? Making voice AI as reliable as any other software you ship.

Thanks for taking a look. Please give us an upvote and share your feedback. Really curious to know what you think!

Try Simulate -> https://shorturl.at/EOLeD

For more details, check out our docs -> https://shorturl.at/WKghM

We started using Future AGI three months ago while chasing that elusive “five-nines” accuracy claim. So far it’s delivered exactly what it promised: the eval suite caught a 4% drift in our image-classification model over a weekend, and the observability traces pointed us to a single batch-norm layer that was mis-scaled on the new GPU nodes. We hit 98.7 % within two days of tuning. The only hiccup has been that alert volume is high out of the box—filtering took a bit of work—so a curated “starter rule-set” would save new teams time.