Ibrokhim Movlonov — Senior Mobile Engineer

Offline-first architecture

ExpoPlatform · 2025 – present

Problem: Exhibition halls are connectivity dead zones, yet attendees rely on the event app for schedules, exhibitor directories, meetings, and lead capture — live data measured in tens of thousands of records per event.
Approach: Own the data layer end-to-end: a delta-sync engine against the platform API, Core Data persistence, and prepopulated per-event databases generated through the app's own write pipeline — so every normalization rule propagates identically to first-launch data and live sync. Real-time updates ride a hardened WebSocket channel.
Impact: Full event content available with zero connectivity across 150+ branded apps built from one codebase; write-side consistency eliminated a whole class of data-mismatch bugs.

SwiftCore DataSQLite WebSocketsSwift Concurrency

Performance

ExpoPlatform · 2025 – present

Problem: On the largest events, exhibitor lists and profiles — the app's most-visited screens — loaded slowly enough that users noticed, with the cost hidden inside broad Core Data fetches and repeated filtering passes.
Approach: Re-architected the data access path: scoped fetches that read only what the screen renders, lazy relationship loading, parallelized reads with structured concurrency (async let), and collapsing multi-pass filtering into a single pass. Locale-aware search and sorting (including Cyrillic) moved down into the database layer.
Impact: Dramatically faster list and profile loads on events with tens of thousands of records add real before/after numbers.

Core Dataasync/awaitInstruments Performance profiling

Cross-platform migration

ExpoPlatform · 2026 – present

Problem: Migrating a mature native platform to React Native (Expo) without losing its hardest-won asset: an offline-first data layer that feature teams could build on from day one.
Approach: Own sync and persistence for the migration — designed the expo-sqlite database schema and the repository contracts that feature teams consume, translating years of native sync semantics (delta updates, conflict handling, normalization) into the new stack.
Impact: Feature teams build against a stable data contract while the platform migrates underneath them — parity with the native engine by design, not by rediscovery.

React NativeExpoexpo-sqlite TypeScriptAPI design

Founding engineer

ONU — AI Health & Prevention · founding engineer · 2026 – present · live on the App Store

Problem: An AI health-prevention startup needed its complete mobile app — architecture, design system, data layer, payments, analytics, localization, and releases — delivered by a founding team of one mobile engineer.
Approach: A modular workspace with clean-architecture modules (domain / data / state / UI kit), BLoC/Cubit state with freezed, ObjectBox offline storage, go_router navigation, RevenueCat subscriptions, AppsFlyer attribution, the Firebase suite, a spreadsheet-driven localization pipeline, and Fastlane match CI for both stores.
Impact: Live on the App Store and shipping regular releases — every layer built and owned by one engineer, which is also the fastest way to prove a stack end-to-end.

FlutterBLoC/CubitObjectBox go_routerRevenueCatFastlane

On-device AI

Apple (via Turing) · 2024

Problem: Apple Intelligence needed on-device LLMs trained and validated at scale, under Apple's privacy constraints, inside a 100+ engineer cross-functional program.
Approach: Contributed to training, evaluation, and optimization of on-device models; built internal tools and automated workflows for data processing, annotation, and model validation, working directly with Apple's iOS, ML, and infrastructure teams.
Impact: Faster annotation and validation cycles for intelligent features shipping in the iOS ecosystem, with privacy-preserving computation throughout.

On-device LLMsModel evaluation Tooling & automationPrivacy engineering

AI tooling

ExpoPlatform · hackathon, team of 3 · 2026

Problem: Cross-platform feature specs were written by hand twice — once per platform — drifting out of sync and consuming senior engineers' time before a line of code was written.
Approach: An agentic CI pipeline: Claude Code CLI on a self-hosted runner inside Bitbucket Pipelines, connected to Jira and Confluence MCP servers, with scoped read-only tokens for both mobile repos. A phase-gated agent prompt (iterated v1→v3) reads the full ticket — description, comments, linked designs, access rules — analyzes both codebases, and enforces strict anti-hallucination rules: never invent class names or endpoints, flag gaps as "needs input from team."
Impact: One ticket ID in — a complete cross-platform architecture spec published to Confluence, with progress reported back to the ticket. Spec-writing collapses from hours to a pipeline run, and both platforms start from the same document.

Claude CodeMCPAgentic workflows Bitbucket PipelinesPrompt engineering

From a founding team of one to Apple's 100+ engineer AI program.