New products and MVPs
Defining the first scope, architecture, and technical foundation. Building the first working version of a product, including a native Android application or a cross-platform solution for Android and iOS.
Deesoft
Applications, automation, and AI-powered tools
I help build new products and improve existing ones: mobile apps for Android and iOS, Python tooling and automation, and AI/ML-powered workflows. I can help define an MVP, integrate systems, diagnose a technical problem, or modernize an existing product in stages.
Tell me about your project
Briefly describe the current stage of the project and what you need.
Scope
What I can help with
Development, integrations, and modernization
New features and fixes in existing codebases, API integrations, module and contract design, and architecture improvements. Also staged migrations to Compose and coroutines, dependency upgrades, builds, tests, and diagnosis of difficult issues.
Tooling, automation, and AI
Python scripts and tooling, data and API integrations, automation of repetitive processes, and AI/ML-powered solutions. The scope may include a prototype, a data processing pipeline, model evaluation, or a tool supporting a specific workflow.
How I work
How I usually get started
- 01
Assess the current state
Whether it is a new product, an existing codebase, a technical problem, an automation task, an integration, or an MVP stage.
- 02
Clarify the missing details
The scope, repository, logs, API, UI design, timeline, and what defines completion for the stage.
- 03
Propose the first useful step
A code review, diagnosis, small fix, MVP plan, technical prototype, or a bounded stage of work.
- 04
Complete a concrete stage
The result may be a working change, diagnosis, plan for the next steps, or technical prototype. I can also continue with code and PR reviews, architecture decisions, and release support.
Technical scope
Technologies and areas I work with
- Android, Kotlin, Jetpack Compose
- KMP and Compose Multiplatform - Android and iOS
- Flutter and Dart - Android and iOS
- Application architecture, modularization, Koin, ViewModel, StateFlow
- REST APIs and backend integrations
- Gradle, builds, dependencies, testing, stabilization
- Local data, synchronization, SQLDelight, SQLite
- Python: tooling, automation, and AI/ML
Selected work
Selected projects and areas of work
Two examples from production applications and two personal projects.
Android - staged modernization
Modernizing an application for a retail chain
This example covers a back-office application used by employees of a large retail chain in physical stores. It supports everyday store operations and integrates closely with an RFID scanner, connecting actions on the mobile device with product scanning. The modernization covers migrating the UI from XML to Jetpack Compose one screen at a time and replacing RxJava streams with coroutines and Flow. The work also organizes UI state around ViewModel and StateFlow within an architecture spanning dozens of Gradle modules.
Android and iOS - cross-platform development
Multiplatform development for retail applications
The work covers internal applications supporting store employees and customer-facing applications for a large retail chain. Two cross-platform approaches are used in production: KMP with Compose Multiplatform and Flutter, both targeting Android and iOS. In existing Kotlin codebases, KMP is introduced gradually by extracting shared logic, sharing UI in selected areas, and integrating new modules with native code. The sharing boundary is chosen separately for each area of an application.
Personal project
Radar: public procurement search engine
Radar is a search engine that aggregates public procurement notices from several independent sources. Its AI pipeline works in stages: retrieving source material, extracting structured fields, validating citations, and completing missing data. Every field carries an exact source excerpt, and a record is stored only after all validation checks succeed. Enrichment fills empty fields only. Quality is measured against golden test sets using tests that replay the entire pipeline and enforce hard acceptance thresholds. The cost of every model call is tracked, while batch processing and model selection help control the budget.
Personal projects
ML/RL agents for card games
These projects cover AI agents for three card games: Pan, Tysiąc (Thousand), and Hearts. Custom training environments were built for Pan and Hearts, while a complete three-player implementation of Tysiąc was added to RLCard. Experiments include self-play PPO, Deep Monte Carlo, and MCTS. Models are evaluated in automated tournaments with Elo ratings and exported from PyTorch through ONNX to TensorFlow Lite with output consistency checks.
Contact