Portrait of Natan Loterio

Natan Loterio

Senior / Staff Android Engineer · Kotlin, Compose, on-device AI · 15 years

Available now. Open to relocation across the EU and UK, with a Luxembourgish nationality application filed and under review.

What I do

I take systems shaped by someone else's constraints and put them back under the team's control. An Android app owned by an outside agency, brought in-house with a team built around it. A fifteen-minute pipeline cut to two and a half. A language model that only ran in someone else's cloud, running entirely on the phone instead.

The work is usually the same shape: find what the team doesn't actually own, and change that. Fifteen years of it, most of them on Android.

Selected work

HUGO BOSS — Tech Lead, Global Android Apps

2025–2026 · HUGO BOSS · 15 countries, 8 languages

Constraint
Move a live commerce app off an external agency without pausing delivery in fifteen markets.
Decision
Insource the codebase as it stood, build the internal Android team from zero, and write the conventions down so a new team could move without a senior engineer in every conversation.
Result
Development brought in-house; the conventions defined for Android were adopted across the whole mobile organization.

Read the HUGO BOSS case study · App

FARFETCH — Android Engineer

2021–2025 · luxury fashion marketplace

Constraint
Modernize a production e-commerce app incrementally, with old and new patterns alive in the same codebase.
Decision
Jetpack Compose adopted screen by screen behind the existing navigation, inversion of control through dependency injection, and the build modularized so Gradle's build cache and configuration cache could turn most CI runs into partial builds.
Result
Pipeline from 15 minutes to 2.5. Compose and DI became the default for new work.

Read the FARFETCH case study · App

Pitaia — CTO

2026–present · AI-native commerce

Constraint
Build a commerce platform where an autonomous agent, not a dashboard, is the product.
Decision
A 26-tool MCP layer exposing the whole product to language models, so the platform is fully operable by an agent.
Result
Blank repo to live stores with paying customers in under six months, including the first Instagram DM agent in the Brazilian creator market to close checkout — conversation to Pix or instalments — without the customer leaving the DM.

pitaia.me

Built in public

Brownie — on-device AI assistant for Android

Kotlin · Jetpack Compose · LiteRT-LM · open source

Constraint
A capable tool-calling agent inside a phone's memory budget, with no cloud and no API key.
Decision
Gemma 4 E2B through LiteRT-LM with quantization chosen at runtime from device RAM, split across five Gradle modules so the inference layer can be swapped when the next on-device model lands.
Result
120 Kotlin files across five modules, multi-turn tool calling, persistent memory, MCP client. Nothing leaves the device.

Read the Brownie case study · Source

Writing

Track record

BSc Computer Science, FURB — Universidade de Blumenau.

Contact