I'm Md. Emamul Mursalin, a React Native developer in Bangladesh who builds AI-powered mobile apps with Expo for iOS and Android. If your app needs a chat assistant, voice input, camera understanding, personalised recommendations or an on-device model that works offline, and you want one developer who can ship both the app and the backend behind it, that is the work I do from Rajshahi for clients around the world.
Most React Native developers in Bangladesh come from a web background and treat the phone as a small browser. Mobile AI is different: latency is visible, API keys cannot live on the device, models have to be chosen for battery and memory, and the App Store review process has opinions about all of it. I build with those constraints from the first screen.
Md. Emamul Mursalin · Rajshahi, Bangladesh · remote worldwide · UTC+6 · Updated
Expo is my default. It gives one TypeScript codebase for iOS and Android, EAS Build and Submit for the stores, and over-the-air updates so a prompt or UI fix ships without waiting for review. On top of it I add the AI layer: streaming chat with the Vercel AI SDK, speech-to-text and text-to-speech for voice assistants, camera and image understanding, and hybrid on-device plus cloud inference where privacy or offline use matters.
Behind the app there is always a backend I build for you — usually NestJS on PostgreSQL — that holds the API keys, meters usage, caches expensive calls and keeps AI costs predictable. Subscriptions run through RevenueCat, StoreKit and Play Billing, and analytics and crash reporting are wired in before launch, not after.
Where the model runs is the first decision. Cloud models from OpenAI, Anthropic or Groq give the best quality and are right for most features. On-device models are right for transcription, classification, quick replies and anything that must work on a plane or never leave the phone. I usually build a small router that picks per request, so the app degrades gracefully when the network does.
Cost is the second decision. A mobile app can generate thousands of AI calls a day from users who pay a few dollars a month, so I design for it: caching, short prompts, cheaper models for cheap tasks, per-user limits, and a usage dashboard you can watch. Every package includes a running-cost estimate before you commit.
Iqtidah is a daily Sunnah tracker and prayer companion app I designed and shipped end to end, with a NestJS and PostgreSQL backend behind it. Virtual Client is a multi-tenant AI voice coaching platform where real-time speech, LLM feedback and tenant isolation had to work together. CakeOrFake is an AI-powered screening tool with subscription billing through RevenueCat and Stripe. The patterns — voice pipelines, metered AI, subscriptions — carry straight into a React Native app.
Starter is $799 over three to four weeks: up to six screens, one AI feature such as chat, voice or image understanding, a backend for AI calls, push notifications, auth and submission to both stores. Professional is $2,299 over six to nine weeks: up to fifteen screens, multiple AI features, streaming and background processing, in-app subscriptions, offline-first sync, analytics, OTA updates and CI/CD. Enterprise is $3,799 over ten to sixteen weeks: up to thirty screens, one AI feature that runs on the device and works offline, company SSO sign-in, encrypted on-device storage and a security review.
Agencies in Bangladesh quote $5,000-15,000 for a comparable app and US studios rarely start below $40,000 once AI is involved. Hourly work on an existing React Native codebase is $20-30 an hour.
I work in UTC+6 with a fixed weekly call inside your working day, TestFlight and internal-testing builds every week so you can hold the app in your hand as it grows, and a shared board for decisions. Payment is by milestone in USD via Wise, Payoneer or bank transfer, the code lives in your GitHub organisation, and the Apple and Google developer accounts are yours. Book a free 30-minute call and bring the app idea; I will tell you which features belong in version one.
Proof
Projects built with the same stack and patterns.
Personal product · Islamic Lifestyle / Habit Tracking
The first Islamic habit tracker that unlocks 50 daily sunnahs by your prayer times.
Robbie Boyle · EdTech / Corporate Training
Teams rehearse live conversations against AI personas and get rubric-scored feedback.
Ayana · HR Tech / Recruitment Automation
AI-powered Chrome Extension that turns hours of Upwork applicant screening into minutes
“This project was handled in an excellent way. The seller understood the task clearly. Progress updates were helpful. The final output was accurate and polished. Everything arrived as promised. Glad to recommend this seller.”
Alexi Denise
Individual client
“He successfully completed all my tasks quickly and with excellent quality.”
Imran Mohammed
“Great work. Great communication, and a quick turnaround. I'd definitely order again.”
Agrossen
Fixed price
Fixed price, fixed scope. Starter and Enterprise tiers are on the pricing page.
Full-featured AI app with subscriptions and analytics
$2,299/project
Launch offer$1,49435% off for the first 5 clients
6-9 weeks
RAG over your data plus automated workflows
$1,299/project
Launch offer$84435% off for the first 5 clients
3-5 weeks
Launch-ready MVP with payments and AI built in
$1,999/project
Launch offer$1,29935% off for the first 5 clients
4-6 weeks
Case studies with the same stack and patterns.
Costs, architecture and how-to guides from the blog.
Questions & answers
My fixed prices are $799 for a Starter app with one AI feature, $2,299 for a Professional app with multiple AI features and subscriptions, and $3,799 for an Enterprise app with on-device inference and enterprise security. Bangladesh agencies quote $5,000-15,000 for similar scope; US studios usually start at $40,000.
Yes, for most products. One codebase covers iOS and Android, the Vercel AI SDK handles streaming, Expo gives you camera and microphone access, and OTA updates ship prompt changes without store review. Native code is only needed for heavy on-device inference or specialised hardware.
Yes, within limits. Small language and speech models run on modern phones with ExecuTorch, llama.rn or Core ML, which I use for transcription, classification and private features in the Enterprise tier. Larger models stay in the cloud, and a router decides per request.
Expo, almost always. EAS Build handles native builds and store submission, config plugins cover most native modules, and OTA updates are built in. I only go bare when a project needs a native module Expo cannot wrap, which is rare in 2026.
Three to four weeks for a Starter app with one AI feature in both stores, six to nine weeks for a Professional app with subscriptions and several AI features. Store review adds a few days at the end, which I plan for.
API keys stay on a backend I build, never on the device. That backend caches responses, uses cheaper models for simple tasks, enforces per-user limits and reports usage on a dashboard. Every proposal includes a running-cost estimate for the expected user count.
Next step
A 30-minute call is enough to tell you which package fits, what I would cut, and what it will cost to run. No pitch, no obligation.
Or email hello@mursalinsdesk.com
