Conversation intelligence

Every store has a receipt that walks out.

Our AI listens to every conversation on your floor and surfaces the sale that almost happened — the customer, the item, the rupee value — while staff can still bring it back.

Backed by

The Monday-morning view

Money at risk, named by store, named by staff.

app.ostronaut.ai/today Live

Web dashboard · live in our Mumbai retail deployment

What customers say

We were optimising for design. Ostronaut showed us that parents on our showroom floor are asking about safety certifications and after-sales support before they ever ask about colour. We rewrote our customer-conversation playbook in a week.
Founding team, Boingg D2C kids’ furniture · offline showrooms
Every support call used to die in a ticket. Ostronaut turned them into a feature backlog — and surfaced a reseller-intelligence layer we didn’t know we had. Nobody on our team had to retag anything.
Founding team, zop.dev Developer tools · cloud automation

Your business has been deaf.

Every customer-facing team has thousands of human-to-human conversations every week, and almost none of them reach the rest of the business. Ostronaut turns those conversations into the intelligence, insights, and analytics you’re currently flying blind without.

Production scale · last 30 days

3,510,920
transcribed turns
163,800
customer conversations
426,160
coachable moments surfaced
₹2.30 Cr
lost revenue caught

8,400+ hours of real customer audio · Hindi · Marathi · English · code-mixed · live across our Mumbai retail deployment.

Hardware

Three devices. One counter, one rep, one view.

No app install, no behaviour change from staff. The hardware does the work — ambient capture, always on.

All three feed our own Indic AI stack — purpose-built for speaker identification, voice fingerprinting, and code-switching across Marathi, Hindi, and Hinglish. Our models find revenue leaks, compliance gaps, and coaching moments — then turn each one into a personalised practice rep for the staff member who lived it.

Capture → AI → Training. One loop.

Think Gong for the physical retail floor.

Full hardware specs →

Hear it work.

One Mumbai pharmacy counter, twenty-four seconds at a time. Raw audio, then the same clip after our pipeline, then the business events that fall out — each anchored to the exact line that triggered it. The fourth clip is a different counter on a different day, the kind our own audio classifier tags ‘unusable’ — listen to what our transcription stack still pulls out.

1 · Raw capture
What the device hears

Open retail counter. Multiple speakers, code-mixed Hindi · Marathi · English, ambient footfall — the noise floor every off-the-shelf transcription chokes on.

2 · Cleaned + diarised
Our audio + diarisation stack

The same clip, after our pipeline. Conversation reconstructed from the signal. Staff and customer separated turn by turn. No enrolment, no manual labelling.

Customer आपका जेनेरिक होगा ना, मैडम?
Staff जेनेरिक नहीं है। मतलब टॉप कंपनी का है। ये जो कंपनी ने जो बेंटाकन वगैरह बना दिया है ना? मतलब बॉडी में उनके भी मैन्युफैक्चरर रहते है — टॉप फोर मैन्युफैक्चरर है।
Staff बस ये है की बस वो एम आर नहीं रखा है। एडवर्टाइजमेंट नहीं करती। पिछले प्राइस का।
Customer कल का भी पेमेंट — वो जो ऑर्डर प्लेस है, उसका भी पेमेंट अवलेबल नहीं।
3 · Business events
What the business actually needs

Every turn evaluated against our retail conversation taxonomy. Coachable moments surface with the line that triggered them. Confidence-gated — uncertain detections never reach the inbox.

Customer आपका जेनेरिक होगा ना, मैडम?
Staff जेनेरिक नहीं है। मतलब टॉप कंपनी का है। ये जो कंपनी ने जो बेंटाकन वगैरह बना दिया है ना? मतलब बॉडी में उनके भी मैन्युफैक्चरर रहते है — टॉप फोर मैन्युफैक्चरर है।
4 · Tough audio, clean transcript
When the audio classifier gives up

A different conversation. Our own audio-quality classifier flagged this clip ‘unusable’. Our transcription stack came back with seventy-one turns at ninety-five-percent average confidence — pure Devanagari, no transliteration, no language toggle.

Customer इन्वेस्टमेंट नहीं है ना पूरा।
Customer वन टाइम इन्वेस्टमेंट है।
Staff हां।
Customer तभी देखो इतना सैलून हो गया है।

What we built

Our own audio, transcription, and event extraction stacks. End to end.

Six stages, one loop. Each conversation flows from raw audio at the counter to a coachable moment in someone’s queue — without anybody typing anything. The depth of what’s underneath each stage is the reason it actually works in a real Mumbai pharmacy at 3pm. Read the deeper take →

01 Capture Rover device, far-field mic

Always-on capture across the counter. No app, no login, no behaviour change from staff. Audio is uploaded the moment the device sees a network.

What we built

  • Hardware designed for the noise floor of an Indian retail counter
  • Per-device calibration tuned to its own acoustic environment
02 Assemble Conversation windows

Continuous audio is sliced into discrete customer-staff conversations by pause and proximity — not by uniform clock windows.

What we built

  • Conversations reconstructed from the signal, not from clock buckets
  • Same-counter continuity across shifts and devices
03 Transcribe Hindi · Marathi · English · code-mixed

Devanagari is preserved. Multilingual code-switching inside a single utterance is handled in one pass. No language toggle.

What we built

  • Code-mixed Hindi · Marathi · English in a single pass
  • Devanagari preserved exactly as spoken, never transliterated
  • Our own multi-stage transcription pipeline — purpose-built for Indian retail
04 Diarise Staff vs. customer, per turn

Each turn is attributed to the right person. Staff are recognised across conversations and shifts. Customers stay anonymous.

What we built

  • Our own diarisation stack — built for noisy, code-mixed, multi-speaker counters
  • Staff recognised without onboarding rituals or enrolment
  • Accuracy gains we don’t see anywhere else in the market
05 Extract Business events — not topics

Stockout uncoached, bounce unhandled, substitution offered, prescription check missed. Coachable moments, with evidence quoted from the transcript.

What we built

  • A retail taxonomy of moments that matter — not a generic topic cloud
  • Every moment ships with the exact line of conversation that triggered it
  • Lost-revenue valued against the real POS history of the customer
  • We don’t ship a moment to your team unless we’re sure
06 Coach Inbox → practice → measure

Every event becomes a remediation step for the right staff member. The same loop measures whether next month’s conversations got better.

What we built

  • Per-event remediation paths into practice scenarios
  • Same evaluator on practice and production — one standard
  • Skill drift tracked across staff, stores, and cohorts

One shift · one counter

15 minutes at one counter. 5 revenue leaks. ₹3,700 at risk.

See how Ostronaut surfaced every miss — the customer, the item, the rupee value — on a single counter at a leading pharmacy chain in Mumbai. Full transcript, audio, gloss, and uncertainty markers.

See the full conversation →

In stealth until today.

Running across teams building some of the largest healthcare, D2C, and B2B operations in the region.

Narayana One Health

Sales intelligence across their hospital network — who closed, who deflected, why.

Boingg

D2C customer conversation capture — every inbound call turned into product and CX signal.

Zeno Health

Full retail + voice loop across 10 pharmacies — staff coaching from real customer conversations.

BridgeBlitz

B2B sales calls captured + summarised, every staff conversation reviewable.

zop.dev

Customer support calls turned into product signal + reseller intelligence.

Founders

Talvinder Singh & Meera Khokhani.

Four-time founder (OYO Flagship hypergrowth, Tushky · 500 Startups 2013, Pragmatic Leaders · YC W21) and ex-NSDC under India’s Skill India Mission — 10,000+ professionals coached on the practice loop since 2018. That evidence base sits under Ostronaut’s coaching layer. Read the full lineage →

Where this is heading

The moment a customer walks out is the moment you can still save them.

CAPTURE INTELLIGENCE TRAINING devices + audio events + insights coaching + practice one continuous autonomous loop
01 · See

The miss, while it’s still hot.

The customer, the item, the rupee value at risk — surfaced before they’re back in their car.

02 · Save

The recovery, queued up.

A message draft ready to send the moment that stock arrives. One tap from the manager.

03 · Steer

The shelf, what to restock.

What customers are asking for that your inventory misses — sorted by rupees, every morning.

Keep exploring

Two ways to start.