Skip to content
About Speaksy

We build the voice India's businesses deserve.

Global voice AI was priced in dollars and trained on American accents. We started Speaksy to flip both — state-of-the-art conversation technology, built in India, priced for India, fluent in India.

14languages in production
1L+calls handled monthly
<800msconversation latency
₹3.99per live minute, from
The Story

From a mispronounced number to a lakh+ calls a month.

2024

The itch

Our founders watch a Pune NBFC pay dollar rates for a voice bot that couldn't say 'साडेचार हजार' correctly. The question forms: why is Indian voice AI priced and built for America?

Early 2025

First stack

Speaksy's streaming pipeline hits sub-second round trips on real Exotel trunks. First Hinglish model trained on consented, PII-scrubbed Indian call audio.

Mid 2025

First customers

A Mumbai collections team and a Hyderabad EdTech go live. The Marathi/Telugu campaigns outperform their human benchmarks in week two.

Late 2025

The graph engine

Prompt spaghetti replaced by versioned dialogue graphs. Human handoff ships — AI does volume, humans close. All 10 gallery languages reach production quality.

2026

Scale

1L+ calls a month across BFSI, D2C, EdTech, healthcare and logistics. Same mission, bigger phone bill: built in India, beats the world.

What we believe

Four convictions we hire, build and price by.

01

Bharat-first, not Bharat-later

Hindi and Tamil ship with the same care as Hinglish. The next 50 crore customers speak neither Silicon Valley's language nor its price points.

02

The phone call is sacred

For most of India, the call is still where trust forms and money moves. We treat every second of latency and every mispronounced name as a broken promise.

03

Frugal engineering, premium output

World-class doesn't have to mean dollar-priced. We optimize relentlessly so ₹3.99 buys what others sell for ₹12.

04

Radically auditable

Deterministic graphs, immutable versions, 100% call QA. If you can't show a regulator exactly what your AI said, you shouldn't be dialing.

The Team

Operators, linguists and latency obsessives.

A team that has run dialer floors, trained speech models and argued about Marathi honorifics — sometimes in the same meeting.

MK

Manav Khanna

Co-founder & CEO

Ex-collections ops. Ran 400-seat dialer floors before deciding to automate his old job properly.

SN

Shruti Nair

Co-founder & CTO

Speech systems engineer. Obsessive about the last 100 milliseconds and the first three seconds of a call.

AR

Aniruddh Rao

Voice ML Lead

Trains the models that make 'नमस्कार' sound like home in nine scripts.

KD

Ketaki Deshmukh

Head of Linguistics

Guardian of honorifics, dialects and everything global platforms get wrong about Indian speech.

NI

Nandini Iyer

Compliance & Trust

Makes TRAI windows, DND scrubbing and DPDP residency non-negotiable platform defaults.

RT

Ravi Teja

Solutions

Turns 'our telecallers say this' into graph agents that say it better, in 10 languages.

Free 100 minutes · No card required · Zero setup fee

Hear it on your number.
Then decide.

A 20-minute conversation. Your use case. Your language. A live call — not a slide deck.