Mahbub Hassan

About

AI product engineer. I build voice agents, and the products around them.

The short version

I'm an AI product engineer. For the past 2+ years I've worked in AI and the backend behind it: voice agents, private document search, and the systems that keep them running. That sits on 7+ years of shipping mobile and web apps. The longer story is below, if you have five minutes.

Twenty cents of internet

I grew up in Jhenaidah, a town in western Bangladesh, in a house full of newspapers. My father is a journalist, so reading was simply what we did at home.

The internet reached me through a basic multimedia phone on 2G, pay as you go. My daily allowance was about twenty cents, and most of it went into recharging my SIM. When the balance ran out, so did the internet. So I learned to plan ahead: I went to free public Wi-Fi, downloaded tutorials, and watched them later, offline.

Tech blogs became my school. I read everything I could find about programming and hacking, and somewhere in there I started to dream about writing my own software.

Then my mother bought me a computer with money borrowed from a bank. Later she borrowed again so I could upgrade it. With a machine of my own, I took everything apart to see how it worked. I rooted Android phones more times than I can count, and I installed Windows on people's computers around town for some pocket money.

By Class 11 I knew I wanted to build software, so I studied Computer Science and Engineering, and spent most of university building apps the coursework never asked for.

From hobby to job

I started in 2016 as an Android developer. I was drawn to Android by its developer community, and later by clean-code ideas from Uncle Bob. Over the next years I shipped 20+ mobile and web apps in Flutter, native Android, and React Native, for couriers, restaurants, clinics, and the tawk.to live-chat app. I worked on Ratio Launcher, which passed a million downloads, and I led a team of ten mobile engineers.

Then, in August 2024, my day job turned to AI and the backend. We were building a document management system, and I set up RAG and intelligent indexing so people could find and ask about any file without opening it.

The documents that couldn't leave the building

My CEO asked me for something simple to say and hard to build: a platform where nobody has to read every document by hand.

It was for corporate service providers in the UAE, the firms that set up companies for other people. Their work runs on paperwork. Licenses, passports, company papers. Each file needed a manual review, and each customer needed someone to explain the next step. He wanted a product that did both jobs: read the documents, and guide people through the setup the way a patient person would. The documents were sensitive, so they couldn't go to an outside AI service.

So I built it. Mostly alone for the first three months, then with one backend engineer I guided. In about six months, ezyglobal.ai had:

  • Nora, a voice guide that answers from our own knowledge base and walks people through each step
  • a self-hosted Llama 3 model, so no document ever left our servers
  • an AI data room built to cut manual document review
  • the web app, sign-in, and Stripe payments around it

Three parts fought back. Getting the voice model onto the GPU so Nora replied without an awkward pause. Teaching it when a person had really stopped talking. Moving heavy document work into background queues so the app never froze. None of them worked well the first time. I kept at each one until it did.

What I took from it: writing code is the easy part. The hard part is connecting every piece so the whole thing feels close to human, and people get guided through without ever noticing the system underneath.

Now

The screen at the end of every AI pipeline still matters to me, because of all those years in mobile. Today I'm building Mama Desk, a clinic front desk that talks on fully local models and answers in about a second. I also evaluate code written by frontier AI models for xAI and Google. Both keep teaching me the same thing: time every step, and don't trust output you haven't checked.

I enjoy small product teams where one engineer can own an AI feature from the first customer conversation to production.

Elsewhere

My full work history is on LinkedIn, and my code is on GitHub. You can email me at [email protected].