I m Somil, I work on voice agents at Vaani in Bengaluru.
Something I read this week bugged me. One benchmark across 28 voice agents found the median one did no better than a plain words only pipeline. Meanwhile every new duplex model promises it hears emotion, sarcasm and hesitation that a transcript throws away.
Both can be true, but it made me wonder how much of what we call feels more human is just lower latency and better interruptions, and how much is the model understanding the caller better.
From my side, speech to speech models still seem less smart than the text LLMs they re built on. For anything involving tools, account lookups or a long conversation, I d still take the smarter model over the more natural one.
So for people who ve tried both:
Hey Product Hunt 👋
I’m Chandra from Vaani, a small voice AI team in Bengaluru.
Voice AI started with great models and only later got applications, so putting an agent on a real phone line still means stitching together speech, an LLM, telephony, tools and retries. We built an internal tool to configure agents without all that glue. Founders, PMs and developers who saw it kept asking for access, so today we’re opening it up.
What you can do, no code:
Design the call flow, including how the agent handles objections, interruptions and edge cases
Give it actions: hold, schedule a callback, hand off to another agent or a human
Connect a knowledge base so it answers from your docs mid-call
Pick a number and go live
After every call, get structured data: extracted fields and a disposition, so you don’t have to read transcripts to know what happened
It also orchestrates multiple models and switches to a backup if one fails mid-call, so one bad response doesn’t wreck the conversation.
We started with languages where data is scarce, which is why “sounds local” matters so much to us. It’s built to work for your customers wherever they are.
What’s the first voice agent you’d build, and what would you expect to break it?
We’ll be in the comments all day.
Try out the platform first-hand at https://platform.vaani.dev/
Grab 1000 free credits by using the code PHL1000X
@chandrashekran_y How well does Vaani handle interruption and backgrounds noise during real phone conversation????
@amrita_prajapati2 haha, this is a classic problem buddy, more our agents naturally handle, but better, we hand out these control and control to the users, you can control exactly how the agents need to behave.
also, regarding handling background noises, we have build propriety modules that solves for these conditions., would for you to try out and let us know!
@chandrashekran_y @amrita_prajapati2 actually these are essentials for the voice agents to conquer real world interactions out there. Vaani handles it by using multiple layers of noise cancellation models and vaani has its own interruption handling model specially tuned for noisy background. It works like charm. Please try and let us know your feedback.
@chandrashekran_y This is interesting, Chandra. I can imagine how much work goes into getting all those pieces to work together reliably, especially on real phone calls. I like the idea of being able to configure the entire call flow without having to build all the underlying integrations yourself. I'll definitely explore the platform and give it a try. Congrats on the launch 👏
when the agent calls to an iphone user the screening assistant may respond to the call , leading to a agent to agent conversation and unnecessary cost.
Can we handle this in vaani
@shahid_zaman_a nice one buddy, we actually have cracked it hehe, test it out and let me know how it goes
(Quick secret: We used the oldest trick in the trade to solve it, if you figured what it is, do tell me., )
@chandrashekran_y lemme try the platform
@shahid_zaman_a Absolutely, would love for you to check it out!!!
Don't forget to grab the free credits specific to PH users, the coupon code is PHL1000X
@shahid_zaman_a this is very recent feature we added in vaani platform. Vaani talks its way through call screener - try the platform and experience it first hand.
Let me know if you need further help.
Congratulations Vaani team on the launch. The product looks really promising. I want to know what all use case does the Vaani agent covers also what all integrations does Vaani currently provides.
@vishal_goswami7 Hey Vishal, thanks for the wishes!
We already have close to 150 prebuilt templates - across multiple usecases and languages, so go the templates sections > filter by industry (we have close to 10 of them), and find the best template for yourself!
And reg, the integrations, our platform support a lot of integrations, but you wanna make a custom integration., that's also very much possible!
Do try out these features and let us know how it went!
Can't wait to hear from you!
@vishal_goswami7 this platform is pretty validated for multiple vertical usecases like real estate, bfsi, hr tech, healthcare etc - it's pretty easy to build using teacher ai or template.
Very impressive. I created an account and played a little with it.
I've noticed that while Hindi is offered as a default language, all English agents come with only one accent. Are you planning to expand that to various accents?
@ilyasb Happy that you created an account and tried out the platform and you found it impressive!
Yes, we currently have these accents out if the box, also if you wanna explore more langues/ accents, try out the voices on Voice lab, also you can clone your voice!
We are also constantly improving our library with more n better voices.
@ilyasb thanks for trying the platform, glad you liked it.
Congratulations on launch! I have two questions
How does it handle interruptions from the speaker?
How does it handle cases where the other side is an IVR or even more challenging, another AI?
@rishi_patel11 thanks for the wishes!,
Means a lot for us!
Coming to your questions,
Interuptions, we have a inhouse propreitory models that take of the interupptions, more than that, based on the use case and also the customer's pace and everything, the agent during the call automaticaly adjusts and calibrates is pace and interruption sensitivity,
2. We already have systems in place for voice mail detection and how you'd like the agent to detect n react to the voice mail
Vanni works for the vernacular languages too?
@avirup_sarker Absolutely, Low resource languages is one of our biggest strengths, would love for you to try it and let us know how it goes!
The handoff to a person is the bit I'd spend time on. The caller shouldn't have to repeat the details they already gave the agent.
@oleksii_sekundant Absolutely, all agents prior to the call handover to the human agent will deliver a summary of all the key points of the calls.. This way the caller never have to repeat anyhing, the human agent can take over from where the agent left!
@oleksii_sekundant exactly, and this is pretty much configurable as well in platform. You can set in the number of calls for which it should maintain memory and it remembers more accurately than human will ever do. Thanks for trying our platform