Mritunjoy Das (Joey Dash)

Worked, studied and got picked at

IIT Madras2016 – 21
ISRONESAC intern
Société Générale2021 – 23
Google AppscaleTop 100 · 2023
Razorpay RizeCohort
Nirmaan IITMPre-incubation
Bengaluru Tech SummitAI partner · 2026
Bajaj Finserv AI FestShowcase · 2026
India Book of Records2017

I build the models and the money rails: speech and decision models that run collections, and fintech that 400K people have used.

users in Subspace's first 12 months: 400K+
processed through Subspace: ₹100 Cr+
model-led collection calls in one deployment: 4.54 lakh
payers committed to pay in that deployment: 26,962

Works.

Vocallabs AI. Foundation models for finance, trained on the lender's own data.

#FoundationModels#Speech#BFSI

Vocallabs AIFoundation models · BFSI2025 – now

I co-founded Vocallabs to build our own models for finance: Crisp for speech recognition and synthesis, and NoDues for collection decisions. Lenders run them inside their own walls, on their own call data.

Role: Co-founder

Showcased at Bajaj Finserv's Finserv Intelligence AI Fest, and named Strategic AI Partner for the AI and Robotics Pavilion at Bengaluru Tech Summit 2026.

Learn more

Crisp.

Our own text-to-speech and speech recognition behind one API, so voices and accuracy are ours to tune.

NoDues.

A collections model that decides who to call, when, and what to offer, trained on the lender's own call data.

Phone and WhatsApp.

SIP trunking and WhatsApp Calling, so the models reach people on whichever line they pick up.

Inside your walls.

Runs in the lender's own cloud or on-prem, so models and call data never leave their perimeter.

NoDues. An agentic OS for collections that decides whom to pursue, when, and how.

#Fintech#Lending#DecisionModels

NoDuesCollections · Lending2026

NoDues is the decision layer for collections. It scores every account, decides whom to pursue, when and how, then runs that plan across voice, WhatsApp, SMS and email.

Role: Co-founder · Vocallabs AI

One civic deployment: 4.54 lakh calls across 64 wards of Bengaluru, and 26,962 property owners committed to pay.

Intelligent targeting.

Scores every account on behaviour, risk and payment signals, so effort goes where it recovers money.

Decision engine.

Your collection strategy written as policies, run the same way across millions of accounts.

Omnichannel.

Voice, WhatsApp, SMS and email run as one plan per account, not separate blasts.

Human in the loop.

Runs on its own and hands an account to your team only when it needs a person.

Subspace. Making the money friends share boring again.

#Fintech#Consumer#Mobile

SubspaceFintech · Consumer2023 – now

I founded Subspace for the money that friends, flatmates and teams share: split bills, shared subscriptions and group savings, in one app on Android, iOS and the web.

Role: Founder and CEO

400K users within 12 months and ₹100 Cr+ processed, with very little spent on marketing.

Learn more

Split anything.

Trips, rent and dinners split in a tap, with a running balance for every friend.

Shared subscriptions.

Groups share streaming and software plans, and everyone pays their part automatically.

Wallet and bill pay.

A wallet, gift cards and BBPS bill payments, in the same app as your groups.

Picked early.

One of India's top 100 apps in Google's Appscale Academy 2023, and part of Razorpay Rize and Nirmaan.

AgentPay. Payment rails for the agentic era, so an AI agent can pay without a form.

#Payments#AgenticCommerce#API

AgentPayPayments · Agentic commerce2026

AgentPay lets AI agents, chat apps and voice assistants take payments safely. A person grants a mandate once, the agent spends inside it, and every charge carries the intent behind it.

Role: Founder · Subspace

Part of the same bet as Subspace: agentic AI that runs payments and commerce, not just conversations.

Mandates.

Authority granted once: scoped, capped, time-boxed, and revocable in one call.

Intent-aware payments.

Every charge records what the agent was asked for and why, so disputes stop being guesswork.

Real-time authorization.

The mandate is checked at the moment of purchase, with device, velocity and amount signals scored in flight.

Built for agents.

Idempotent endpoints, typed SDKs and an MCP server, so a payment is a function, not a form.

Quietly. Taking my YouTube feed back from the algorithm.

#AI#OpenSource#ChromeExtension

QuietlyChrome extension · MIT2026

My YouTube home feed had turned into Shorts and game clips. I built Quietly so a small decision model judges each video once and only the useful ones stay.

Role: Solo builder

About $0.0002 per 40 videos, live on the Chrome Web Store, and a launch post that reached 77K+ people.

View code

Judged once.

Each video gets one verdict and it's remembered, so the feed never asks the model twice.

No flash of junk.

New tiles stay hidden until they're judged, so nothing you'd skip ever shows up first.

Replies in your voice.

Suggests replies on WhatsApp Web and Gmail in your own style. You always press send.

Your key, no server.

Runs on your own OpenRouter key. Only titles and channel names leave the browser.

Grow90. A free school and project hub for engineering students across India.

#Community#Education#NonProfit

Grow90Non-profit · EdTech2020 – 21

I started Grow90 at IIT Madras as a non-profit, data-driven consulting community. It grew into free courses for students and real projects for developers.

Role: Founder

10,000+ students taught for free and 1,500 developers working on projects for startups and MSMEs.

Free courses.

Bootcamps and webinars on ML, deep learning, competitive programming and system design.

1,500 developers.

A place for student developers to pick up real projects across many domains.

Accomox.

A flexible e-commerce module, picked for the Nirmaan IIT Madras pre-incubation cohort.

Data for MSMEs.

We used 2 million users' data to help startups such as Techtable and Edar Labs grow.

From the lab.

CFI, IIT Madras · 2017 – 18

Flight control board

A low-cost flight controller running a PID loop on a custom PCB, at half the usual cost. Funded by IC&SR, IIT Madras.

Open source

Joeygql

GraphQL building blocks with two-factor auth, Razorpay and Instamojo payments, mail and SMS. 1,500+ users, including IIT Madras's internship fair.

ISRO (NESAC) · 2019 – 20

Object tracking on a drone

SiamMask and YOLOv3 ported to an NVIDIA Jetson TX2, tracking objects in real time on board.

IIT Madras · 2017

Multi video chat

Group video calls over the campus intranet with WebRTC, open sourced for every student.

Inter-IIT Tech Meet · 2018

Drone flips

An algorithm that flips a quadcopter n times mid-air without losing altitude, for the PlutoX problem.

I also flew a microlight, ran Saarang's tech, and still write code most days.

I studied aerospace at IIT Madras and started out on flight controllers and drone autopilots. Then I moved to software: risk systems that certify a million trades a day at Société Générale, a group-finance app, and now foundation models for finance.

I like owning a problem end to end: talking to users, writing the backend, and shipping the thing on a Friday. Bengaluru is home.

Journey

2025 – nowCo-founderVocallabs AI
2023 – nowFounder and CEOSubspace
2021 – 23Risk teamSociété Générale, Bengaluru
2020 – 21FounderGrow90
2018 – 20InternshipsISRO (NESAC), Société Générale, Gramophone, Detect Technologies, Perpule
2016 – 21B.Tech + M.Tech, AerospaceIIT Madras

My stack.

Build

Go
Python
TypeScript
Dart
Flutter
React
Next.js
Node.js
FastAPI

Models

Speech recognition
Text-to-speech
Decision models
LLM agents
PyTorch
TensorFlow
OpenCV

Fintech

BBPS
KYC
Wallets
Payment mandates
Collections
SIP telephony

Ship

PostgreSQL
Hasura
GraphQL
Kubernetes
Docker
Google Cloud
AWS
n8n

Building in lending, payments or collections?

Write to me about foundation models for finance, or whatever you're building.

Book a call LinkedIn