Hamming AI offers automated testing for voice agents

Here's an interesting one if you're working with Conversational AI Voice Agents: Hamming AI. The startup is offering automated testing for voice agents to flag quality issues in development and production. I caught their Y-Combinator launch post recently and thought I needed to document it here.

How do you test your Voice Agents at the moment?

I trust the answer isn't, "By phoning them and seeing what they're like!"

Here's the workflow that most teams follow (according to the Hamming team):

1. Call your voice agent by hand and find bugs. Slow and ad-hoc.
2. Tweak your voice agents by adding new tools and changing the prompts or models to fix the bugs.
3. Call again to see if the changes worked.
4. Detect regressions when users complain of things breaking in production.
5. Repeat steps 1 to 4 until you get tired.

Does anyone recognise this? ;-)

Hamming has a different approach.

They create loads - hundreds - of characters to simulate real-world interactions with your voice agents. Then they unleash these characters on your Voice Agent systems - and mark the results.

One of my favourite Hamming features is this one:

When customers complain about a bad call, you can locate the call transcript and create a new character in one click. Make a change to your prompt, and then run the simulations to ensure you addressed the bad call.

If you're interested to learn more, perhaps one of the next steps is to check out this demo video by co-founder Sumanyu Sharma:

One of the user feedback comments I saw on the Hamming.ai website describes Hamming.ai like this: "This is like Optimizely for building AI products" - I like that!

I don't see specifics on the business model but I'd imagine it's something like a per-call or per-usage approach. The team suggest that their AI voice testing is, "1000x faster and 20x cheaper than using humans."

If you want to try it out yourself, right now, you can:

Here's the link to their voice demo.

Good luck Team Hamming.ai!

You can contact the team directly via their website or find them on LinkedIn here: Samanyu Sharma and Marius Buleandra.