Yelaman Yelmurat

Key contributions

• Led modularization of a legacy mobile app into 14 business domains, establishing Domain-Driven Design (DDD) and Clean Architecture boundaries. Each domain encapsulates its models, repositories, state, and UI, enabling parallel development streams, reducing merge conflicts, and improving incremental build times.
• Adopted Claude as the team AI assistant and rolled out AgentSync, a self-authored open-source tool, so the team keeps one .ai/src source for 13 AI coding tools, cutting estimated hands-on time per shared-rule update from ~20-40 to ~2-5 minutes (~75-90%).
• Designed an update distribution and remote configuration system combining native update mechanisms (Google Play In-App Updates, iTunes Lookup API), Remote Config (Forced Update, Feature Toggles resolved per flavor, platform, and minimum version, with route-level gating and placeholder screens), and Shorebird Code Push, enabling Dart-layer hotfixes in minutes instead of waiting through the standard 1-7 day store review cycle.
• Rebuilt mobile CI/CD with dependency caching and parallel iOS/Android pipelines, cutting build times by ~60% (>10 to ~4 minutes) while automating TestFlight and Firebase App Distribution delivery, notifications, release diffs, and Jira ticket linking.
• Built merge-blocking CI gates for static analysis, tests, formatting, localization, shell tests, and AI-config drift after the infrastructure lost its macOS runner, cutting pipeline runtime 23% (25.5 to 19.5 minutes) through two-layer caching of the pub cache and the build_runner asset graph, which removed ~250 s of cache re-archiving per pipeline and let code generation run incrementally. Fixed a bash version difference that let test assertions pass unnoticed on macOS.
• Implemented digital signature flow via government PKI service (Deep Linking, OAuth 2.0 via WebView, web fallback) that replaced a desktop signing agent. Cut signing time from 5-10 minutes per document to ~1.5 minutes.
• Cut application size by ~26% total (35 MB to ~26 MB) in two stages. Wrote a custom WebP asset compression script that reduced the app from 35 MB to 29 MB, then authored a Python scanner that detected and removed unreferenced assets across all sources, freeing an additional ~3 MB.
• Standardized integration and review practices across the 4-engineer team through trunk-based development and pair review. Created Individual Development Plans for 2 engineers to formalize technical-growth goals.
• Drove technical hiring for the mobile team: conducted 20+ technical interviews covering live coding, architecture, and problem-solving to evaluate Flutter candidates and shape team composition.
• Eliminated production OAuth session races through generation-scoped transactions, stale-callback rejection, serialized secure storage with deterministic iOS Keychain recovery, and HTTPS-only traffic. After release, reported authentication failures stopped recurring.
• Reduced reported analyzer findings by 99% (~2,000 to ~20) through code cleanup and analysis-configuration changes. Removed unused imports, fixed controller and notifier disposal, and configured strict cast, inference, and raw-type checks.
• Built a design system with 50+ reusable widgets and a shared GLSL shimmer clock, reducing 400+ per-widget tickers to 1 on skeleton-heavy screens.
• Integrated interactive 3D models through WebGL (model-viewer) into the mobile app for product and event showcases.
• Filtered ~85% of bot registrations by integrating reCAPTCHA Enterprise into the mobile registration flow, obtaining action-specific tokens and attaching them to registration requests for backend verification.
• Added screen capture protection to sensitive QR, ticket, and quiz screens through a shared lifecycle-aware component that disables capture on entry and restores it on navigation or disposal.
• Built a privacy-safe support code that links mobile errors to backend X-Request-ID logs and the matching Sentry event. The QR and text payload stays under 60 characters and contains no credentials or PII. During a production incident, the team used the code to find the cause in a few minutes. Based on its manual steps, the old process was estimated to take tens of minutes.
• Automated analytics event integration with analytics_gen for a contract of 42 domains and 306 typed events, cutting setup from 10-15 minutes to 2 through typed code and documentation generation, with CI publishing schema changes to analysts and product through Telegram.
• Automated updates of 50-100 Figma design tokens by building Flart, a TypeScript plugin that reads Figma variables and generates typed Dart classes and ThemeExtension themes. Replaced manual copying and comparison, reducing transfer time from ~1 hour to minutes while keeping colors, typography, and variables consistent with code.
• Architected a multi-backend layer with isolated clients for internal services and payment gateways (Kaspi Pay, Freedom Pay). Decoupled base URLs, interceptors, and error handling reduced new API integration setup from ~1 day of boilerplate to a few hours.
• Shipped a modular SQLite data layer using Drift ORM, caching 18 entity types across 14 feature modules, with atomic transactions, stale-while-revalidate reads, and hybrid TTL/row-count cleanup. Cached data is served immediately and refreshed in the background after TTL expiry. Offloaded large-dataset parsing to Isolates to prevent frame drops during list scrolling on poor networks.
• Queued moderator QR scans in a SQLite write-side queue and synced them on typed triggers (app start, connectivity restored, scanner opened, app resumed, backoff elapsed, manual retry), keeping rate-limited scans pending and pacing the queue by the server's Retry-After, so event check-in no longer depended on momentary venue connectivity.
• Migrated dependency injection across 14 feature modules from Service Locator (GetIt) to Pure DI with a manual composition root and constructor injection. This turned missing dependencies into compile-time errors instead of runtime exceptions and made dependency graphs explicit and testable.
• Established a three-language localization workflow (~2,100 keys per locale) for an outsourced translation team and delegated the first implementation to a mid-level engineer. During review, fixed issues with key ownership, overwrites, metadata, and validation, then rewrote the Python scripts in Dart. The final workflow protects structural Google Sheets columns and checks ARB files in CI for invalid JSON, mismatched ICU placeholders, empty values, and missing CLDR plural forms; the validators depend only on dart: libraries, so the gate runs before dependency installation. These checks catch structural translation errors before merge instead of after someone sees them in the app.
• Initiated Astana Hub's Anthropic Certified Partner enrollment. Drove the company application and coordinated 10 engineers through the Claude Partner Network learning path, qualifying the organization for the Claude Certified Architect B2B tier.
• Delivered the Digital Bridge networking module covering WebSocket chat with auto-reconnect, match cards, QR business-card scanning with contact saving, and two-way device calendar synchronization across iOS and Android. Replaced manual contact exchange and disconnected scheduling with a single in-app flow for connecting and planning sessions.
• Launched Digital Bridge event modules covering ticketing, schedule builder, QR check-in, and AI Quest gamification. Implemented check-in through dynamic QR codes displayed after ticket activation and read by standard scanners. During the event, the flow processed 14,400+ QR scans with 86% ticket activation; the event hosted 27K+ attendees.
• Built alem.llm, an OpenAI-compatible Dart client for the in-house AlemLLM API. It supports chat sessions with managed history and context, streamed responses, and typed domain models that decouple the UI from backend contracts, powering internal LLM-assisted features.
• Rolled out Server-Driven UI (SDUI) for forum-related pages where content changes frequently, moving layout and content control to the backend so updates ship without an app store release.
• Engineered navigation on Octopus Router with stateful nested stacks, giving each tab independent history, type-safe arguments, and deep-link state restoration. The router reconstructs the active tab, stack depth, and nested screens so users land in the correct context.
• Integrated ISpect diagnostics across HTTP, WebSocket, application state (BLoC), and SQLite (Drift), giving QA shareable in-app access to logs, network traffic, database activity, and device state.
• Stopped cross-line duplicate build numbers by syncing remote Git state before every version bump and selecting the next number across all branches after a real collision. Each release record stores flavor-scoped first-parent history and one entry per delivered platform, covered by 98 Bats tests for rebases, rejected pushes, rollback, and notification failures, so every build is traceable to its source commits.
• Fixed unreliable returns from the government app by saving the original route, arguments, signing flow, and request ID before handoff. The app restores only the matching transaction and deletes it after use, so users return to the right screen and the app polls the right request even after Android killed the process during signing.
• Diagnosed a cross-platform eGov Mobile login failure to seven distinct root causes, from an OAuth future severed by Android task switching to a Dio content-type conflict that failed before the request left the device. Fixed the client-side causes with regression tests and documented a diagnostic decision tree that isolated the final server-side blocker for the backend team, whose fix restored Android eGov login.
• Hardened Android releases for 16 KB memory pages by building an APK/ELF alignment checker and upgrading the NDK and packaging configuration to verify native libraries before distribution.
• Stabilized API behavior after server rate limits exposed redundant request bursts. Eliminated unnecessary calls and implemented bounded HTTP 429 retries with Retry-After, exponential backoff, jitter, and per-request opt-out.
• Aligned the CTO and backend engineers on a shared mobile API direction by using list-heavy screens to demonstrate coupling and over-fetching, and proposing versioned endpoints with a mobile Backend-for-Frontend contract to decouple the client from service-internal contracts.
• Instrumented behavioral analytics, attribution, and experimentation with Microsoft Clarity session insights and heatmaps, AppsFlyer install attribution and campaign measurement, and A/B testing. This gave marketing channel-level acquisition data and enabled feature validation against real user behavior before full rollout.
• Replaced manual spreadsheet collection with in-app surveys that load server-defined questions, capture typed choice, rating, and text responses, and submit structured answers through a REST API. This gave the product team structured per-segment feedback from visitors, speakers, and partners for the first time, informing future event planning.

Key contributions

• Cut app startup time by ~60% (from ~3.0s to ~1.2s), measured from main() to first interactive screen. Offloaded heavy synchronous operations from main(), implemented parallel async dependency initialization, and added lazy-loading of modules.
• Designed a feature-first architecture with explicit BLoC state management and repository boundaries. Covered ~80% of business logic across BLoCs, repositories, and use cases with unit tests, verified through CI coverage reports.
• Cut application size by ~35% (50 MB to 32 MB, Play Console data) by removing unused dependencies and legacy assets, lowering storage requirements for budget devices.
• Built the core EDM module for document creation, approval, and resolutions, with access controlled through ACL role models. Integrated government PKI signing by sending requests to a server-side signing service for user confirmation.
• Moved vacation requests from physical HR visits into the application by implementing an in-app request workflow.
• Cut diagnosis time for network and system errors from hours of log-digging to minutes by implementing structured logging and filtered search.
• Set up push notification delivery across Google GMS and Huawei HMS, covering every device type in the enterprise client base.
• Developed multi-step authentication using SMS/email one-time passwords (OTP), PIN, and Secure Storage. Integrated Google ML Kit face detection to check face presence and position in the login camera frame.
• Moved file viewing, uploads, storage, and version history from the web portal into the mobile Drive module through a typed HTTP client (Chopper) for a dedicated file-service API.
• Built background GPS tracking through native platform integration (Method Channels), balancing location accuracy with power consumption for field monitoring. Tuned update intervals, accuracy, and distance thresholds, adjusted updates to movement, and limited tracking to the active work scenario.
• Reduced repeated onboarding explanations by creating the team's first Architectural Decision Records (ADR) and guidelines, giving new developers a shared reference they could consult independently.

Key contributions

• Optimized AgroGIS rendering of 5,000+ objects via dynamic zoom-level clustering, lazy loading (on-screen items only), and Isolate-based parsing of large datasets and charts. Eliminated the 2-3s freeze during zoom/pan on budget Android devices.
• Built offline-first architecture with OpenStreetMap tile caching, SQLite persistence (Drift), and conflict resolution with update prioritization. Defined API contracts and typed clients with Retrofit, aligning sync and error handling with the backend to maintain cache consistency.