Hey, I'm Dmytro 👋

Get stable and scalable mobile apps

If you're ready to grow your business with a mobile app that's fast, stable, and private — you're in the right place. A decade of Android engineering, 30+ apps shipped, millions of users served, and talks at Cloudflare, MongoDB, and Google Developer Groups.

Book a free call
Dmytro Samoilov
30+Projects developed
12Happy clients
MillionsOf users served
−76%AI cost, cloud-only → hybrid

Seen atCloudflareGDG LisbonMongoDBCritical SoftwareUpwork Top 1%

How can I help?

Building something in health, fintech, messaging, or anything EU-regulated? Pick the problem you walked in with:

What clients say

Straight from Upwork and LinkedIn, unedited.

IP
Igor P. LOGITY TECH
★★★★★
"Dmytro conducted an audit of our Android application — code quality, architecture, potential issues and others. We received a detailed report with specific recommendations. A year and a half later, we continue applying his advice: improved DI structure, reduced technical debt, implemented static analysis. Professional approach and recommendations that actually work in practice."
KF
Kevin F. Nelu LLC
★★★★★
"Dmytro is an incredibly talented Android developer."
AM
Andrew M. SHEQSY
★★★★★
"Great developer and a team player. Highly recommend"
AH
Andrii H. Private client
★★★★★
"Great android developer! Thanks a lot for your job"
SC
Stanley C. Private client
★★★★★
"Great job! Everything works excellent, task was done very quickly! Thanks a lot!"

Case studies & guides

Real numbers from shipping on-device AI on Android. Written for the person making the call.

Talks

Sponsored by

Constraints of On‑Device AI — What I Hit Shipping a 100% Local Note‑Taking App

GDG Lisbon · Critical Software · Lisbon, Portugal

The numbers behind Offhand: Whisper for speech, Gemma for structure, nothing leaves the phone. I'll show benchmarks from three devices, the thermal throttling that made iteration five 47% slower, and the model-loading waits nobody warns you about.

RSVP on GDG
Dmytro Samoilov presenting device RAM limits for local LLMs on Android at the Cloudflare Lisbon office

Hybrid AI Architecture for Android: LLM On‑Device + Cloudflare Workers AI + MongoDB Atlas

Cloudflare Lisbon Office · Lisbon, Portugal

One AI feature, two backends: Gemma running locally on phones that can handle it, Cloudflare Workers AI catching the rest. Live demo on real hardware, with a Samsung flagship doing speech-to-text fully on-device next to a Pixel 4a routed to the cloud. Same experience, different backend.

Invite me to speak

Work with me

I'm taking on a few companies that want help shipping private, on-device AI. Three ways to start:

Feasibility audit

Start here. A fixed-scope review of your app, your device base, and your data constraints. What you get:

  • Analysis of your codebase, feature set, and real device base
  • A clear call per feature: on-device, hybrid, or cloud
  • Cost math: local models vs cloud APIs at your scale
  • Device targets: RAM floor, chipsets, what to test on
  • The answers your security review will ask for

Fixed scope · roadmap in 2 weeks

Team workshop

One or two days with your Android team, hands-on, in your codebase. When I leave, the knowledge stays.

1–2 days · on-site or remote

Hands-on integration

I build the feature with you: model selection, device-capability routing, fallback rules, benchmarks on real hardware.

Embedded with your team

General Android development

Not every project is an AI project. Features, refactors, performance, releases — senior Android help where your team needs it.

Fractional · ongoing or project-based

Not sure whether on-device is even worth it for your app? Book a free 15 minutes and bring your app and your AI question. If I can't help, I'll say so.

Book a call

Recent projects

Flagship · Live on Google Play

Offhand

Voice notes and meetings turned into clean, structured notes by AI that runs entirely on the phone: Whisper plus a local LLM, offline and encrypted. No cloud, no account, free and open source. It's the same architecture I build for clients, shipped as my own product.

Meet Offhand
Offhand recording a voice note, transcribed 100% offline with on-device speech-to-text

ALAI Bench in development

A benchmark app for Android and iOS that puts open-weight models through real scenarios (audio, text, images) and measures performance and thermal throttling across chipsets. Open source, going public later this year.

Follow me on GitHub →

Claude Code template for Android

Open-source template for using Claude Code with Android projects. Configs, prompts, and workflow automation.

View on GitHub →

On-device image generation (iOS)

iOS app using Apple Intelligence to generate images locally. No cloud API, everything happens on the device.

View on App Store →

Alarmaze — Solve to Stop

A puzzle alarm clock for heavy sleepers: the alarm won't stop until you solve a maze, so your half-asleep brain has to engage. No snooze, no muscle memory. Live Activities and Dynamic Island support on iOS.

View on App Store →

About

Live demo comparing on-device speech-to-text on a Samsung flagship with cloud fallback on a Google Pixel 4a Conversations with attendees after the talk at the Cloudflare Lisbon office

I've spent a decade building Android apps, from architecture to performance. Before that I was the kid flashing custom ROMs: modded Symbian at 15, ported Android builds across devices on 4PDA before I ever wrote production code. Since then I've led a team of 7, went freelance, made it to top 1% on Upwork with 100% job success.

Now I work on the two questions behind every AI feature: what does it cost per user, and where does the data go? Cloud inference is a bill that grows exactly when your app succeeds. And for health, fintech, anything with GDPR in the room, the data question decides whether the feature ships at all.

My answer is on-device first. Gemini Nano, Gemma, Whisper, hybrid on-device + cloud setups: I test them on real devices, benchmark what actually works, and publish everything. Not theory. Real code, real numbers.

Based in Lisbon.

Watch the build

I test on-device AI on real Android projects and film what I find. The parts that worked and the parts that didn't.

Building an AI feature that can't leak user data?

Tell me what you're building and I'll tell you what can run on-device, what needs hybrid, and what it would cost at your scale. Write wherever is easier for you. I read everything myself.