The self-learning voice for finance

Voice agents trained on your calls, checked on every turn, improving every week under your rules — in Hindi, Hinglish, Tamil and more.

RBI · IRDAI · DPDP mappedRuns in your VPC or on-premBorn from IIT research
Live call · hi-IN · collections
R
Riya · FormantAI agentOutbound · Hinglish · EMI reminder
01:14
00:04Agent
Namaste, main Riya bol rahi hoon, [Lender] ki taraf se. Kya main Sunil ji se baat kar rahi hoon?नमस्ते, मैं रिया बोल रही हूँ… क्या मैं सुनील जी से बात कर रही हूँ?
00:11Customer
Haan bol raha hoon. Kya hua?
00:14Agent
Aapki 5 tareekh ki EMI ka payment abhi tak nahi aaya. Kya aap aaj UPI se pay kar payenge, ya main payment link WhatsApp par bhej doon?आपकी 5 तारीख की EMI का पेमेंट अभी तक नहीं आया…
identity_verified ✓intent: promise_to_paytone: bucket-1 ✓
Release · agent v42
$ formant release --agent collections-hi --candidate v42
graded 48,210 calls · 14 failure clusters
regression 1,240/1,240 · frozen lines verified
approver: client QA lead · canary 5% → 25% → 100%
live v42 · rollback ready
Trusted by
IndusInd Bank
Banking · India
TeamLease
Workforce operations · India
Google for Startups
Programme
IIT research
Origin
Speech

It speaks like your best agent.

Fine-tuned on your own recordings — accent, pace, numbers and dates the way your customers say them.

Reasoning

It checks every turn.

Grounded in account data, your script and your regulator's rules before a word is spoken.

Learning

It gets better every week.

Graded calls become the next release — regression-tested, human-approved, one-click rollback.

Ownership

It stays inside your walls.

Your model, your keys, your region. India cloud, private VPC or on-prem GPUs.

Most voice bots are as good on day 300 as they were on day 1. Ours run a governed weekly learning cycle on your own calls.

The engine

Every turn checked before it is spoken.

live call · hi-IN · collections · agent v42 live
R
Riya · FormantAI agentOutbound · Hinglish · EMI reminder
01:14
00:04Agent
Namaste, main Riya bol rahi hoon, [Lender] ki taraf se. Kya main Sunil ji se baat kar rahi hoon?नमस्ते, मैं रिया बोल रही हूँ… क्या मैं सुनील जी से बात कर रही हूँ?
00:11Customer
Haan bol raha hoon. Kya hua?
00:14Agent
Aapki 5 tareekh ki EMI ka payment abhi tak nahi aaya. Kya aap aaj UPI se pay kar payenge, ya main payment link WhatsApp par bhej doon?आपकी 5 तारीख की EMI का पेमेंट अभी तक नहीं आया…
00:27Customer
Salary Friday ko aayegi, tab kar dunga.
00:31Agent
Theek hai Sunil ji, main Friday, 18 September ka promise note kar rahi hoon. Link bhi bhej rahi hoon. Shukriya.ठीक है, मैं शुक्रवार 18 सितंबर का प्रॉमिस नोट कर रही हूँ।
stage: identity_verified ✓intent: promise_to_payPTP: 2026-09-18tone: bucket-1 · no pressure ✓PII masked ████
reasoning trace
00:04ai_disclosureverbatim · frozen
00:11identity_verifiedname match
00:14amount_lookuplms.due=₹4,850
00:14policy_checkrbi_fpc · tone=bucket-1
00:31ptp_captured2026-09-18
00:33qa_grade94 / 100
stage graph
ptp dispute paid low_confidence → re-ask Greeting lang_detect · consent Verify identity name · DOB / last-4 Classify intent policy_check · amounts Capture PTP date · channel · link Raise dispute ticket → CRM Confirm & close UTR lookup · thanks Every turn passes three checks before it is spoken ✓ grounded in account data (no invented amounts) ✓ script & regulator rules (RBI FPC calling hours, tone) ✓ language & persona consistency (frozen closing lines) Warm transfer → human anger / legal threat

Inside the multi-stage engine

Recursive learning loops

Every call makes the next one better.

A turn check inside a call grade inside a weekly release inside a portfolio — four loops, each feeding the one outside it.

01 · Listen

Every call, every outcome, every correction

Recordings, transcripts, customer outcomes (PTP kept, payment received, policy renewed), human-agent takeovers and compliance flags stream in as signals — 100% of calls, not a sample.

02 · Grade

Every call scored on your rubric

Grounding, flow, compliance, tone. Failures are clustered: "repeated the EMI amount", "missed already-paid", "asked branch-or-link twice".

03 · Curate

Redact first, then choose the lessons

Aadhaar, PAN, card and phone numbers masked in text and audio. Your best agents' calls and the corrected failures become the training set — approved by your reviewer.

04 · Fine-tune

Your model, your data only

Speech and language models fine-tuned per tenant. No shared brain. The weights are yours, stored in your environment for VPC and on-prem.

05 · Simulate

Proven before it speaks

The candidate replays 1,240 golden scenarios — barge-in, angry customer, already-paid, wrong number — with Hindi/English parity and frozen-line checks. Any drop blocks release.

06 · Release

A human signs. Canary. One-click rollback

Named approval from your QA or compliance owner, staged rollout 5% → 25% → 100%, automatic rollback on a guardrail breach.

07 · Measure

Outcomes, not minutes

Containment, promise-to-pay, right-party contact, complaints and cost per resolved call — tracked per release, reported every morning. Then the next week begins.

swipe → 7 stages
01
Listen
calls · outcomes · corrections
48,210 signals · 7d
02
Grade
every call scored on rubric
100% graded · 14 clusters
03
Curate
redact · select · approve set
PII 100% · 3,120 examples
04
Fine-tune
your model, your data only
candidate v42
05
Simulate
golden set · barge-in · parity
1,240 scenarios · 0 red
06
Release
human approval · canary · rollback
5% → 25% → 100% · 1-click back
07
Measure
PTP · containment · complaints
PTP +8.5 pts over 6 wks
next week's calls feed the next cycle · client memory: scripts · glossary · frozen lines · escalation matrix
Compliance lines never learnHumans approve every releaseRegression before releaseOne-click rollback

The full learning cycle

Evaluation models

The models that grade the models.

A second model, trained on your QA team's labels, scores every call on grounding, flow, compliance and persona. It replays 1,240 scenarios against every candidate — and any regression blocks the release. Your reviewers recalibrate it weekly.

Regression scoreboard · v42 vs v41
ScenarioGroundFlowComplyPersonaΔ vs v41
already_paid · hi-IN↑ was red
barge-in during amount
wrong number · Marathi
legal threat → transfer
salary-delay reason code↓ flow 0.82
disclosure line · verbatimfrozen ✓
angry customer · Tamil
branch-or-link asked twiceblocks release
1,240 scenarios · 4 rubrics · 2 languages · any red blocks release illustrative · replace with live scoreboard
Built for regulated finance

Insurance. Banking. Lending.

Collections, renewals, KYC, lead qualification and service — in the languages your customers actually speak.

Deploy your agent

Three ways in. One governed engine.

Pilot · 6 weeks

Start with one journey

One use case, one language, your scripts. Live calls in week 3, learning cycle from week 4, a decision by week 6.

How we work
Platform · your model

Build on the engine

Model Studio fine-tunes speech and reasoning on your recordings. Telephony, WhatsApp, SMS and chat from one brain.

Explore the platform
Enterprise · inside your walls

Own the whole stack

Private VPC or on-prem GPUs, air-gapped option, your keys. Security reviewed against RBI, IRDAI and DPDP.

Deployment options
Our story

From an IIT lab to live calls in five countries.

FormantAI started as speech and reasoning research — the hard problem of understanding Hindi, Hinglish and Tamil the way people actually speak on a phone line. It became a company when the first bank asked us to run it inside their walls.

Research
01

It began in the lab

Speech and language research at IIT, on code-mixed Indian speech over real telephony — the audio most models were never trained on.

Product
02

Then a first live journey

A regulated lender needed a voice agent that could be audited, not just demoed. Every turn checked, every line frozen where the regulator demands it.

Learning
03

Then it learned

Graded calls became weekly releases — loops inside loops, judged by evaluation models we built before the learner. Better in month six than in week one.

Today
04

Now live in five countries

India, UAE, USA, Vietnam and South Africa. Part of Google for Startups. Trusted by IndusInd Bank and TeamLease.

100%
of calls graded, not sampled
<1 s
turn latency on Indian telephony
11
languages, native-trained
1-click
rollback on any release
Pilot

Hear it in your language

Book a demo
Engineering

Scope it with an engineer

Talk to an engineer