Caller
What the customer sees and hears: the orb, the language, the hold music, the close.
A real-time voice agent that answers the phone line in বাংলা or English, knows who is calling, verifies them the way a trained officer would, and resolves requests on live core-banking data, inside the institution’s own guardrails.
White-label · your institution’s name, your assistant’s voice
Banks and financial institutions whose phone line carries more volume than the floor can answer.
Balances, transactions, cards, limits, dues, loans and deposits, from live core-banking data, never from memory.
Only what the institution’s policy allows on a call. Anything outside it is refused, with the real process offered instead.
Every call streams to an operations floor where a supervisor can whisper, take over, or let it close itself.
Nothing about an account is said before the bank agrees the caller is who they claim to be, and nothing is changed that policy has not allowed. The same discipline, on every call.
The line connects securely, the greeting comes in the caller’s language, and a returning customer is recognised before the first word, so the call starts with context rather than a form.
Identity is checked the way a trained officer would, and the decision belongs to the bank’s own systems rather than to the conversation, so the agent cannot hint at an answer or be talked past the check.
Every figure comes live from the core system, never from memory. Every change is checked against the institution’s policy first, and a refusal carries the real process the customer should follow instead.
When a supervisor decides, or the customer asks, an officer joins with a brief of the call so far. Once the call ends, it is quality-reviewed automatically.
What the customer hears, what the floor sees, and the control room behind both. Every surface works from the same live picture of each call, so a figure on the floor and a figure in the transcript are the same figure.
What the customer sees and hears: the orb, the language, the hold music, the close.
The call as it runs, with the transcript and every system call rendered inline.
Live headline figures and a card per call, so a supervisor sees the whole floor at once.
The control room behind the floor: customers, accounts, cards and loans in one view.
Recognises a returning customer, so the call starts with context rather than a form.
Officer-grade identity checks, decided by the bank’s own systems rather than the conversation.
Live answers across accounts, cards, loans and deposits, available only to a verified caller.
The institution’s own product and service knowledge, for the questions that need no account at all.
Every requested change checked against the institution’s own policy before anything happens.
Whisper, take over, queue, hold music and a warm brief for the officer joining.
Every refusal, every attempt and every escalation, timelined against the transcript.
Completed calls scored automatically, so quality is sampled from all of them rather than a few.
Most voice AI fails on the parts nobody writes requirements for: the ringback, the pause before the answer, how a figure is spoken aloud. Those are the parts that decide whether a customer stays on the line.
The whole state of the call, made visible in a single orb: turning slowly while idle, rippling while ringing, swelling with the voice, shifting while thinking, grey when muted.
It greets in the caller’s language and follows them when they switch, mid-conversation. Everyday mixed Bangla and English is understood as naturally as either one.
Ringback while the line opens, a chime on pick-up, hold music through a transfer, a tone when the officer arrives, and a clean close. The familiar cues that tell a customer they are on a real call.
Figures are spoken in lakh and thousand the way people actually say them, and written in lakh grouping on screen. Account and card numbers are never read out in full.
A single animated orb shows the call at a glance: it turns slowly while idle, ripples while ringing, swells with the voice, shifts while thinking, and greys when muted.
Those are two different lists, and the second one is enforced by the bank’s own systems rather than by the conversation. Nothing is read before the caller is verified, and nothing changes that policy has not allowed.
The conversation and the system activity in one column, so a supervisor sees not just what was said but what was looked up, what was refused, and how long each took.
Account data is released only to a verified caller, and that is enforced by the institution’s systems rather than by anything the agent is told. An attempt to get around it is refused, and the supervisor sees it.
The institution’s policy decides what may change on a call, and it leans protective: a lost or compromised card can be blocked the moment it is reported, while riskier requests go to the proper channel.
A refusal is never a dead end. The customer is told what the actual process is: which branch, which form, which channel, so the call still resolves something even when the answer is no.
Balances, dues and limits come live from the core system, never from memory or guesswork, and every lookup is visible to the supervisor in the transcript.
Live headline figures and a card per call. A supervisor can whisper into the agent’s ear without the caller hearing, take the call outright, or let it close itself. Every completed call is graded automatically afterwards.
Live count, verification state, who is waiting for an officer, containment rate, average handle time and guardrail blocks, then one card for each call in progress, with the last thing said on it.
কার্ডটি ব্লক করা হয়েছে, আর ডিসপিউট নম্বর SR-260919-4821। সাত কর্মদিবসের মধ্যে জানানো হবে।
আমি আপনাকে আমাদের ফ্রড টিমের একজন অফিসারের সাথে যুক্ত করে দিচ্ছি। একটু লাইনে থাকুন প্লিজ।
আমার মোবাইল ফোনটা চুরি হয়ে গেছে, অ্যাকাউন্টটা সেফ করতে চাই।
Headline figures across the top, then one card for every call in progress: who, how long, which language, and whether anything has been refused on it yet.
The conversation and the system activity in one column: every lookup and every decision sits between the turns it belongs to, with the full detail a click away.
A supervisor can put a line into the agent’s ear that the caller never hears, or take the call outright, with a warm brief so the customer never has to repeat themselves.
Completed calls are scored automatically, so quality assurance samples the whole day rather than the handful anyone had time to listen to.
White-label from the ground up: the institution’s name in Bangla and English, the app, and the assistant’s name and voice all carry your brand, everywhere a customer meets them.
We will stand the agent up against a sandbox of your core system, load your own products and policies, and let your supervisors watch it work a live floor.