Animesh MahatoApr 9, 2026
Introducing Wavelength: AI Voice Agents for India 🇮🇳
Introducing Wavelength: AI Voice Agents for India 🇮🇳
Animesh Mahato·Apr 9, 2026
Animesh, co-founder of Chanik, introduces Wavelength, an AI voice agent platform designed specifically for the Indian market. He explains how the tool solves common issues like robotic voices, high costs, and poor interruption handling to create human-like conversations.
Chapters
00:00
Introduction
Introduction to Animesh and the motivation behind building Wavelength.
00:41
Market Challenges
02:32
Intelligent Interruption Handling
05:02
Humanizing the Experience
Transcript
00:00
Hi everyone.
00:01
So I'm Animesh and I am the co founder and CTO at Chanik.
00:05
So this video's main purpose is to walk you through a tool we built for our own problems and later on got to know that a lot of businesses need it.
00:15
So we created a product out of it and we are serving people every day with this product.
00:21
So I just want to walk you through the reason behind creating the product and how is it different than everything that is there in the market?
00:29
So the product is called Wavelength and the main thing is it's AI voice agents that actually sound humans and are built for India.
00:38
Let me go in detail.
00:41
So first thing, is like why even create a voice agent platform when there are literally tons of them already in the market, right?
00:51
Big companies, small companies funded, non funded, all the players are there in the market.
00:55
Then why are we entering this space still?
00:59
There were several reasons, big reasons.
01:01
In fact, we at Freedom with AI, so we have a Freedom with AI company also where we train people on how to use AI, which we have trained more than 480,000 people all across the world.
01:13
So we wanted to do AI calling, right?
01:14
And it was having major issues.
01:17
Number one, the voices were very robotic, right?
01:20
They were not at all human sounding.
01:22
You can call it out, it's an AI pretty easily, even if you use one of the most expensive models in the world.
01:28
Now the biggest second problem we faced, especially in India, was it was English only.
01:35
So other languages were there.
01:36
It's not like it was not there.
01:38
But a lot of other languages were not sounding human at all.
01:44
Next, the expenses it was costing all the way from 6 to 13 rupees or 7 cents to 15 cents a minute, which is very unrealistic for India.
01:55
In India you can hire a human at that price or cheaper.
01:58
It's not at all worth to get a lower quality voice at a higher quality price.
02:04
We can't go with that.
02:05
And lastly, it was very sterile, like it was too AI.
02:10
Like it sounds like, okay, this person spoke and it's now silent.
02:14
There is nothing going on in the background.
02:16
You can easily call it out.
02:17
It's an AI.
02:18
So these are some major issues that we saw in all the existing providers that we wanted to use.
02:26
So now that you know all the issues, let me show you what all solutions we came up with.
02:32
So the first feature I'm very proud of is intelligent interruption handling.
02:36
You know how we humans talk, right?
02:38
When I ask you a short question like what is your name?
02:41
You say Animesh.
02:42
And it's a pretty short answer.
02:44
I have to immediately reply, great, my name is Priya.
02:47
Something like this.
02:48
But if I ask you what is the reason that is making you wake up every day and work towards your goals?
02:54
It's a little deep question.
02:55
It takes time to think, right?
02:58
So just imagine you start thinking, okay, I, you know what I think and.
03:02
really?
03:03
What do you think?
03:04
You know, I think getting, up every day and working towards my goal brings me closer to, you know.
03:11
And then AI interrupts again.
03:13
what does it bring closer to?
03:14
Like, this sounds a little jagged, isn't it?
03:18
Like the person on the side is not even a good listener.
03:21
That's the problem with AI, in 90% of the cases, like the problem with AI calling, in fact is latency is the first biggest problem that I, I understood that AI should immediately respond.
03:33
But a bigger problem than that is judgment on when to respond immediately and when to take some time.
03:38
So I literally built an entire infrastructure that intelligently handles interruption.
03:44
Let's say you are saying, the AI is talking in a way, or someone is talking in a way.
03:50
And we say some filler words like Okay, got it.
03:52
It will not take that as a complete input and regenerate the entire sentence.
03:56
Again.
03:57
It can ignore when it's supposed to be ignored.
04:00
It will take an account when it's supposed to take an account.
04:02
For example, AI says, you know what, you can join this master class and level up in AI.
04:09
And then the person says, okay.
04:11
And then the AI will continue to speak instead of saying, yeah, I think you agree.
04:15
Let me say that this will not happen.
04:17
It will ignore the user's, fillers and have a proper conversation like you have with humans.
04:23
Right.
04:24
Next, complex answers.
04:26
It will wait a little longer for you to give a response.
04:30
Like I explained the first example, second, third is a yes or no question.
04:35
It will respond immediately.
04:36
So it feels like it's instant.
04:38
It's not an AI who's thinking for a long time.
04:40
And last is real interruption.
04:43
As I said, whenever you are giving an answer, you're taking pauses, you're taking filler words.
04:48
It accounts for couple of, things and understands when to interrupt, when not to interrupt.
04:53
It's a system that took a lot of fine tuning, but it works pretty good.
04:58
Next, sounds like a person in an office.
05:02
This is something we, just came up with randomly on a Tuesday that can we add a bit of office noise in the background that might make it feel like it's little more human?
05:13
And we added that and actually worked out good.
05:16
In our favorite we added good decent amount of office noise in the background, elevators, keyboards and all.
05:22
And it worked out pretty good.
05:24
Next thing we did is normalization.
05:26
Even with good AI models for voices, AI is mostly trained on extremes sometimes.
05:34
So let's say I'm talking with you.
05:35
Okay.
05:36
Do you know how AI can help you?
05:38
The person answers something and like great.
05:41
This is absolutely right.
05:43
The great should not be so high pitched.
05:45
It should be slightly high pitched but not so much.
05:47
So what we experienced with most of the models we tried using or we are using right now also that some words it's too much over and some words is too much lower.
05:57
So to, to account for that gap, we did normalization that whatever is too loud or little less loud can be normalized to a pace that it still feels like the emotion is behind it and it doesn't feel artificially enhanced kind of an emotion.
06:13
Next we have filler detection.
06:15
I already told in this, the person says hum, and all the fillers in all the Indian languages, we can understand that and we can account for that also.
06:26
Like let's say you are talking, you know what I should learn AI.
06:29
I'm like okay.
06:31
And then the person says yeah, because this is happening, that is happening.
06:35
So it can intelligently put fillers wherever it's needed.
06:40
Next, beyond just making calls.
06:43
So all I spoke in the last five minutes was how the calling works.
06:47
But there is a game after the calling also which is much more important, if not equally important.
06:54
Smart scheduling.
06:55
Lot of times we notice is people will say I'm busy, can you call me back after 10 minutes?
07:00
And a lot of AI systems do not have this automatically scheduling.
07:05
So we implemented the system.
07:07
If you say to our AI I'm busy, can you call me after 5 minutes?
07:10
It will call you exactly after 5 minutes I'm busy, can you call Me up at 10:30?
07:14
It knows current time, it knows when is 10:30 and it will schedule a call for 10:30.
07:18
Right.
07:19
Then call memory.
07:21
So we added the system so that one single bot Persona can talk to a person in multiple checkpoints.
07:28
For example, how we are implementing this in Freedom with AI is when you sign up 5 minutes after sign up we give you a call, we confirm that you're coming and everything.
07:38
We give you some freebies and all for like small consulting call.
07:42
On the day of Master class, we call you again saying hi.
07:45
On that day we talk.
07:46
You remember you told that you're trying to learn AI to not get replaced Yeah, I remember.
07:52
So are you coming to tonight's master class?
07:54
Yes, I'm coming.
07:55
Lets say this person misses the master class.
07:57
Next day the bot calls them again.
08:00
This time.
08:01
Okay, I remember you told me you committed to coming and you also said it's a burning issue because you might lose your job but still you didn't attend.
08:10
What is the reason?
08:11
So it can hyper personalize every single conversation based on the past call data and you have it completely configurable to yourself.
08:19
Next.
08:20
It has 12 plus Indian languages.
08:22
All the mainstream Indian languages are there.
08:24
Hindi, English, Tamil, Telugu, Malayalam, a lot of them.
08:27
Most of the Indian languages are there.
08:29
And fine tuned voices are there for every single language.
08:32
Next, Auto, DNC and follow ups.
08:35
So if someone says just stop calling me, okay?
08:39
or someone is pissed off, someone like I don't want to get any calls from you.
08:43
How do we deal with that?
08:44
We have a smart DNA list which gets populated with these people.
08:49
So even though you schedule a campaign from your CRM from Wavelength anywhere, it will not call them.
08:55
Reason being they are in dnd.
08:57
So it helps you save a lot of compliance headache.
09:01
Next, the bot circuit breaker.
09:03
At the end of the day, even if you fine tune this to a lot, there is of course a chance that the bot can go rogue or your prompt can have issues.
09:12
So, so we have personally experienced it that in some cases the bot is answering in the wrong way or it is having the wrong date and it kept calling hundreds of people.
09:20
So we have bot circuit breakers that can automatically stop the bot from executing calls if it's behaving in the wrong way.
09:29
Next, voicemail and IVR detection to save your telephony cost.
09:33
whenever the call goes into IVR or it goes into voicemail, the bot immediately cuts the call and says I'll call you back later.
09:43
Next, you can connect Wavelength with literally any CRM that you have or any system that you have.
09:51
All it has to do is have some endpoints for the APIs and webhooks.
09:56
After Collins, our AI does its own extraction of the data and scores the lead and it can send them WhatsApp messages, emails and tag in your CRM.
10:05
Do whatever you like in your own CRM so you do not have to migrate completely to a new system.
10:11
If you want to use Wavelength from tomorrow, you can integrate it to absolutely any system you're using.
10:19
Next.
10:20
Best quality at best price.
10:22
So here are some competitors as you can see and their prices.
10:26
Only competitor which is close to us is Bolna or Plevo.
10:29
But in the terms of quality we can definitely say we are better if not as good as them at a competitive price.
10:40
When it comes to western alternatives like bland, warpi and retail, we are much more cheaper than them and we are providing more indianized formats for all the people.
10:52
Next, intelligent analytics and safety.
10:56
So you have completely configurable analytics to yourself.
10:59
You can set up a bot goal.
11:01
So what is the bot's main goal?
11:03
For let's say some bots it could be setting an appointment.
11:06
For some bot, it could be making sure people are coming to the master class.
11:10
For some bot it could be something else.
11:12
So it's completely configurable and you can set up any bot goals and you can also set up red flags.
11:17
For example if you don't want people to feel like it's an AI.
11:22
So you can set an okay, if person asks are you an AI that's a red flag.
11:26
So you have to fix your prompt so you can set up any sort of red flags that you want.
11:31
Someone let's say says the date as wrong date, there is another red flag you want to set up in red flag.
11:36
As soon as that triggers it will automatically pause the bot.
11:40
So it's not hard coded anywhere.
11:42
You can set up your own goals, you can set up your own red flags.
11:47
Next.
11:48
Real impact, real numbers.
11:51
The main thing that we want so for freedom with AI we have done a lot of calls and we have.
11:59
Our main thing is if we get you run ads, we get people to sign up and they show up to the webinar then only we can help them upskill.
12:06
In AI if we are not able to have people show up which is the main goal for creating the wavelength, we are basically losing on our customer acquisition cost.
12:16
So just with using wavelength we increased our show up rate by 20%.
12:20
Previously it was 30%, now it's 50% and we have done 22,000 calls in just the last 35 days.
12:27
So it's not that we are preaching you to use our system, we are using it every day ourselves.
12:32
And you can get it on the get in on the action immediately by seeing the real life impact.
12:39
Next.
12:43
We don't just give you tool like other providers do, we actually help you win with it.
12:49
How?
12:51
We as soon as you onboard give understand your business, design a prompt by understanding your customers you and then make sure you get to your success outcome within a week or two.
13:04
So we help you onboard A to Z including prompts, including setting up bot, including setting up your CRM.
13:10
Once everything is done, your team can take it over and create as many bots as they like.
13:17
Of course, if you are someone who likes to do it all by yourself, you have full independence.
13:21
But if you are someone who's on the edge, okay, I have tried wopi, I have tried millis, I have tried retail.
13:27
I have tried all these tools.
13:28
The biggest problem is I don't know how to prompt.
13:30
I don't know how to set it all up.
13:32
Don't worry.
13:32
Our team will take care of that.
13:34
All you need to do is sign up for Wavelength.
13:37
Okay?
13:38
So I hope this video helped you understand how Wavelength works.
13:42
The next I'm gonna show you a small demo with some analytics on how Wavelength actually looks from inside, so you can actually go about learning how to use this tool.
13:52
And it's stupid simple.
13:54
It literally takes 10 minutes to learn Wavelength, and you can go about using it.
14:00
So, as you can see, I'm logged into our, main Wavelength account, and here we have all the analytics in one single page.
14:08
This is like a small dashboard that you can have to see your analytics.
14:12
We have a detailed analytics page also.
14:14
But for now, this is the setup.
14:16
You can change the range, let's say, in the last seven days, and you can set up specific bots.
14:21
How is everything going on?
14:23
Okay, so this is the base analytics.
14:26
Next we have the bot configurator.
14:28
You can set up your bots here.
14:29
We'll go to this next because this is one of the crucial parts of the system.
14:34
So we'll go to this in detail.
14:36
But before that, I'll show you the rest of the ui, whatever calls you want to manually, trigger.
14:41
You can search the lead here, and you can select the bot, and you can trigger those calls from here itself.
14:47
Okay, this is for your internal testing.
14:48
If your team wants to test any of the bots, they can trigger manual calls from here.
14:53
Then we have the call queue.
14:54
This is where I told you, told you about circuit breakers.
14:57
At any point in time, if you want to pause a bot from doing any calls, you can pause it, and all the calls will be stored in the call queue.
15:06
Okay, next we come to call history.
15:09
You can see all the calls that have been happening in the bot.
15:12
You can filter them by category.
15:14
Let's say I want to see all the completed calls within, let's say, the duration of two to five minutes.
15:20
I can see all of them, right?
15:22
And I can filter by bot.
15:25
I can filter by date.
15:26
I can filter by sentiment, interest level.
15:31
It has red flags or no red flags.
15:33
Which red flags are there, what objectives I want to tackle for.
15:36
All the filters are there.
15:38
Okay.
15:39
And when you open any of these calls, you will see this kind of analytics.
15:43
You can see the recording actually you can play it in any speed.
15:46
You can download the recording, you can see the actual transcript and you can see the summary on the right.
15:51
You can also see the extracted data.
15:53
This is completely configurable by the way.
15:56
For us, we want to know what is the motivation for the person for signing up, how was the call quality and everything?
16:03
Right.
16:03
And next is the talk ratio, how much in the call the bot talked and how much the person talked.
16:09
So this is the base level analytics.
16:10
You can get about any calls that you have.
16:13
It's pretty, pretty detailed.
16:14
You can see the exact transcript, you can see all the details of the customer.
16:18
Next, coming to the detail analytics page.
16:21
You can select whatever bot you like, time range, and you can see all the calls that are happening.
16:26
As you can see, we have a 72% success rate with average call duration of about 2 minutes and red flags of 16%.
16:34
This red flag also include the person being silent so that removing that it would be less than 5 to 6%.
16:41
Now as you can see, you can see recently analyzed calls.
16:45
Also you can see the call quality, how much of the call bot was speaking, how much the person was speaking, the sentiment, and average call duration over time and average word share for every call.
16:59
Next, we have objections if you have certain objections.
17:02
Like for example, if someone is challenging the bot's identity, that is an objection for us.
17:07
So we are highlighting that so we can go listen to that specific call.
17:10
You do not have to dig through a lot of calls to find this objection.
17:13
It comes in one place itself, right?
17:16
Then lead intelligence, how many leads we are getting, is it warm, cold and hot.
17:20
So that helps your ads team or your sales team to understand if the lead quality is an issue or not.
17:26
Then at the end we have the cost, how much cost we are paying per bot conversation.
17:32
Right?
17:32
So this is the data analytics.
17:34
Then we have the leads page where you can see all the leads that are coming in the system.
17:39
And when you open any of the leads, you can see exactly the conversation that happened and all the things.
17:45
For example, if I open my lead, you can see all the calls that have been happened with my contact will come here and I can mark someone as a DND right here only.
17:57
Right?
17:58
So these are the last three calls that have happened to my contact and all the summary of them then is the contact Call history completely.
18:07
So whenever they did not pick up, when they picked up, everything is there sequences are something like.
18:13
Which basically is like sending out certain messages and emails that happen post the call.
18:20
Okay, next we have team management.
18:23
You can add team members as you like.
18:25
And this is how easy it is to learn Wavelength.
18:30
Now let me get to the actual bot configurator.
18:33
I'll have to log into a test account so that I do not reveal the entire prompts and everything.
18:38
So let me log into a bot account and I'll show you how easy it is to set up any bot in less than 10 to 15 minutes.
18:45
So I've logged into a test account.
18:47
I'll go to bots and create a new bot here you can name the bot.
18:52
Say I name it as Priya.
18:53
You can name the company.
18:55
Let's say we name it as Chanik AI.
18:59
This greeting, you can whatever you want to have something.
19:02
Example.
19:03
Hi Animesh, Sneha here.
19:05
Can you hear me?
19:06
I just changed Sneha to Priya.
19:07
Or you can also use agent name variables here if you don't want to hard code the values.
19:13
Second is a callback greeting.
19:15
Let's say if the person already had a call and the next time we are calling them again, we can even modify that greeting.
19:23
Something like hi customer name.
19:24
This is Sneha.
19:25
do you remember we had a call sometime back just like this, Right?
19:34
You can add a location, event name, event date, event time.
19:38
These are variables which are predefined.
19:41
You can of course add your own variables.
19:42
Also when you come to voice, you have speech recognition.
19:46
We have two models, Model one and two.
19:48
You can cycle through them to see what works for us.
19:51
We have seen model two working much good.
19:54
And then we have all the voices that are there.
19:58
Then you can also change languages.
20:00
These are all the languages we support as of now.
20:03
Right?
20:04
And this is something which is an experimental feature we have released recently where you don't mind AI taking a little longer to answer.
20:12
You can turn it on so it will think a little longer.
20:15
It will be a little more smarter to talk to the person.
20:18
But the latency would be little high because it's thinking in the prompt.
20:22
You will actually add your prompt.
20:24
Something like you can simply add as basic as you are a customer support agent calling from Chanik AI and have the rest of it and then it will take the conversation forward.
20:40
Next we have context variables.
20:42
This basically is like boosting certain keywords.
20:45
So let's say you are running an AI company.
20:47
So for you words like chat, GPT, AI, Claude, these words would be common.
20:52
So you can add these.
20:53
This will help the AI to basically learn new variables, right?
20:58
And if you create any new variables, for example test variable, you can define the default value of the test variable Also next we have telephony providers.
21:10
We have partnership with Plevo and Twilio both Plevo for India, Twilio for international.
21:17
If you have any of them, you can start using the app.
21:20
Then we have integrations.
21:22
If you have a go high level CRM you can integrate the CRM directly here.
21:26
And we have three types of triggers.
21:28
Pre call, during call, post call, right.
21:31
Based on any of them you can do tagging.
21:34
So as soon as the call starts you want to add a tag in the CRM you can add during the call.
21:39
For example, if customer is asking to send credentials again I want to add a tag called send credentials and workflow name could be anything simple like send credentials workflow and it will add that tag during the call itself.
22:02
So during the call itself you can send them any communication you like.
22:05
And then we have post call after the call is completed.
22:08
If you want to send them a confirmation after call you can do that.
22:13
So confirmation message, something like this you can set up.
22:17
If you are someone who likes to entertain.
22:19
You can Also connect any 10 automations by having webhooks and you can send any kind of data you want.
22:27
Transcript, bot config details, contact analysis, everything.
22:31
If you want to send, you can send along with this call and do whatever you like with the data.
22:36
Send as many reminders messages as you like.
22:39
Then we have the goal configurator where you can set up certain goals.
22:42
For example what is the primary goal of the bot?
22:44
What is the secondary goal of the bot?
22:46
What are the red flags?
22:48
What are the data capture you want to do for the customer?
22:50
All this you can set up here.
22:52
Next we have settings.
22:54
So what is the max call duration?
22:56
So that no one can abuse your AI.
22:58
If someone just keeps on talking it can run up your bill.
23:00
So we have added hard limit setter.
23:02
You can set this to any amount 8 minutes.
23:05
10 minutes after that the AI will automatically cut the call silence timeout.
23:09
If you are having a conversation with a person who would require a little more time to think, we have a system where it can check in like something like can you hear me?
23:17
It will ask that based on if the customer is silent for very long that timer you can set here.
23:24
So let's say you want to make it say that after 10 seconds of silence.
23:28
Then you can set it up to 10 seconds.
23:30
Then we have max concurrent calls.
23:31
How many people you want to call at a certain time.
23:34
You can set it to any number as you like.
23:36
Then we have circuit breaker, as I already told you, the number of threshold failures that will happen.
23:42
Like number of red flags that happens based on that you can turn on and off the circuit breaker.
23:47
After three times, bot makes mistake, stop the bot.
23:49
Basically that's the use of this.
23:52
Then we have office background noise.
23:54
If you want have the office background noise, you can have it by turn on turning on here and call memory.
24:00
You can set up this as if the board will have X number of calls memory in the context before calling this person.
24:07
Let's say on an average this call goes in the third sequence of the calls.
24:12
So you can set up all the last three calls.
24:14
The last five call memory it should have before calling that person.
24:18
If it already had talked to the person.
24:20
Then sequences are something I already told you about.
24:22
It's in development.
24:23
It will allow Wavelength to send out all the messages and emails directly from Wavelength itself.
24:29
Then we have callbacks.
24:31
If you want to set up something like okay, after the person doesn't pick up, call them after three hours again, then try once again.
24:37
If doesn't pick up the next day in the morning try once.
24:40
The next day after evening try once.
24:42
So we have pre built templates, but it's completely customizable.
24:45
Whatever you want, you can change it as this basically default callback.
24:50
Of course if someone says call me after 10 minutes, it will call them after 10 minutes only.
24:54
Then you can set up the time zone of the bot.
24:56
The calling window also is there for your compliance.
24:59
For context, in India you can call from 9am to 9pm so you can set it up to call from 9am to 9pm Then we have one more thing called bot switching.
25:08
So let's say the bot you designed is for English and the person says I do not understand English, can you talk to me in Hindi?
25:14
You can say yeah sure, my colleague can give you a call back in five minutes.
25:18
let me have another colleague called me back and you can select another bot and that bot will call them in a different language.
25:26
Let's say you want person wants to talk in Hindi, it can turn on the Hindi bot and they'll get a different call from the Hindi bot.
25:32
So this is how you can set up the bots.
25:35
This is like pretty straightforward and one time setup.
25:38
Once you make it, you can just run this bot infinitely so that's how you go about using wavelength.
25:44
And I know there's a lot of features and a lot of things which are Indianized we have created.
25:49
I would appreciate your feedback.
25:50
Whatever you think you can share us in the form below or comment on this video.
25:55
And yeah, until then, take care.
25:57
Bye Bye.
Comments
Comments
No comments yet