We build the models, then we build what runs on them

Tracing back to our Founder's childhood aspirations, he always wanted to build something that learns intuitively, something more than a calculator - something that could solve all the problems he could not find the answers to.

His passion for overcoming language barriers led to assembling a Speech Research team and founding Intelsense AI Research in 2018.

Intelsense AI founded by speech scientists, NLP researchers, software engineers and machine learning experts, to build ASR technology.

The goal was to make speech recognition more accessible and accurate for businesses and committed to facilitating an inclusive world in which people's access to technology is not limited by their native language.

For the first time in the history of Bangladesh, SenseVoice is a transcription platform where any user can input any Bengali or English audio or video file to automatically generate text from the speech found in the uploaded files.

We received feedback from global users including researchers from Google, World Bank, Amazon, and Asian Development Bank. Thousands of members have started to use our products to solve their daily pain points for free.

Intelsense built a speech recognition model small enough to run on a Raspberry Pi Zero or a Pi 3B.

The Bangla Voice Controller Module lets people operate their Walton air conditioners by speaking Bangla to them.

AirVoice is a voice search platform for e-commerce, built for customers who would rather shop by picture and voice command than by typing.

Alongside the e-commerce work, Intelsense is working with private banks to put voice assistants into their operations.

Five years of speech research gave us the acoustics. What Bangla still lacked was a language model that understood it natively rather than as a translation target, and a synthetic voice worth listening to.

Both became products the following year. Building them in house, rather than fine-tuning somebody else's base model, is the decision the rest of the portfolio now rests on.

Ekush LLM launched as the first Bangla large language model built in Bangladesh, covered by The Financial Express, Dhaka Tribune, Prothom Alo and Asian TV.

With SenseSpeech alongside it, the stack was complete: speech in, language in the middle, speech back out. Every product we have shipped since runs on these models.

The models became a workforce. UniSense AI shipped with six agents covering the full customer lifecycle: find, convert, onboard, retain, support and collect. They share one view of the customer, so nothing is lost in a handoff.

It went live across real estate, education and immigration, and internet service providers, which is where the ARR lift figures on this site were measured.

AirVoice became FinSense AI, an agentic operating system for banking that handles both the customer conversation and the paperwork behind it. AIHubspot opened every frontier model on one subscription priced in taka. Altercrew entered early preview as a governed digital back office built with RunAgent.

Four products, each solving a different half of how a business runs, all of them running on the Bangla speech and language stack we spent eight years building.