Software · Embedded · Systems Developer
C/C++ · Python · TypeScript · Kotlin · ESP-IDF · Android · Linux
I build practical systems where software has to deal with real constraints: hardware state, limited resources, unreliable networks, long-running processes, local data, explicit authority and failure recovery.
My work spans embedded firmware, Android, developer infrastructure, local-first products, applied ML and hardware tooling. I prefer explicit contracts, observable state and testable boundaries over hidden behavior.
Android research/product prototype for completing bounded real-world tasks over ordinary cellular calls on a stock Samsung phone. The system combines real call audio, local Polish STT/TTS, on-device Gemma 4 through LiteRT-LM, deterministic dialogue paths and an application-owned authorization layer for external effects.
Engineering signal: real-device telephony · local inference · safety/authority boundaries · Android services · STT/TTS · failure recovery
Stack: Kotlin · Android · Gemma 4 · LiteRT-LM · STT/TTS · Shizuku
Deterministic local execution infrastructure for AI-assisted development. A planner defines exact work; Local Agent executes it locally under explicit time, memory, repository and resource constraints and publishes machine-readable evidence of what actually happened.
It includes multi-repository scheduling, isolated workspaces, durable claims/results, process-group lifecycle control, watchdogs, crash recovery, repository/resource leases, Git-backed control state, emergency controls and macOS launchd deployment.
Stack: Python · Git · process groups · threads · file locking · launchd · GitHub Actions
Local-first Android app for configuring a thermostat-style climate automation without a hub. The phone configures and diagnoses; a Shelly Plug S Gen3 executes the installed automation locally using BLE thermometer data, so the default path does not require Home Assistant, MQTT, cloud services or a 24/7 server.
The project has a verified real-hardware path, published Android beta builds and an accepted UX baseline.
Stack: Kotlin · Android · Shelly RPC/Scripts · BLE/BTHome · local-first automation
Project page: https://michalmatu.github.io/shelly-link/
Native ESP-IDF growbox controller for ESP32-S3 with deterministic climate control, real sensors, RF433 outputs, durable telemetry, e-ink operator UI and TinyML research tooling.
Production control remains deterministic and safety-owned; ML is deliberately kept in a shadow/research role. The firmware includes a native SSD1680 display path, Clay-rendered operator pages, host-side C++ tests and hardware-focused verification flows.
Stack: ESP32-S3 · ESP-IDF · C++17/20 · FreeRTOS · SCD41 · RF433 · e-ink · TinyML
Public showcase for my larger private ESP32-S3 automation platform. GrowClip combines a node-based automation runtime, sensor integrations, BLE/MQTT connectivity, persistent history/archive, an embedded web interface and device-side UI across constrained display targets.
The full firmware/runtime stays private; the public repository contains the marketing site, product mockups and a standalone LiteGraph demo.
Stack: ESP32-S3 · C/C++ · FreeRTOS · BLE · MQTT · microSD · Clay · SvelteKit
Live: https://michalmatu.github.io/growclip-site/
| Project | Focus | Main stack |
|---|---|---|
| BlueEye Tracker | Local-first Bluetooth/BLE situational-awareness app with explainable evidence, bounded ingest, Room persistence and field-tested Android scanning | Kotlin · Jetpack Compose · BLE/Bluetooth · Room · Hilt |
| MatrixHub | ESP32-S3 sensor/display platform with SCD41, BLE, Wi-Fi CSI motion sensing, USB HID, notifications, Shelly integration and embedded web UI | ESP32-S3 · C/C++ · PlatformIO · TinyUSB · SvelteKit |
| Miauudio | Android-first ambient audio mixer with native Media3 background playback, imported audio, procedural generators, PWA/web target and release tooling | TypeScript · React · Capacitor · Kotlin · Media3 |
| PhotoMap | Place-centric visual map platform with editorial media pipeline, moderation, FastAPI backend and React/Leaflet frontend | Python · FastAPI · SQLModel · React · TypeScript · Leaflet |
- Hardware Lab — consolidated firmware starters, board bring-up, ESP32-S2 USB networking, ESP32-C6/Zigbee experiments, Rust/ESP32, nRF52840, Raspberry Pi, e-paper, PCB/SKiDL/KiCad work and computer-vision experiments.
- IleStoi.pl — maintained map application for documenting long-standing vehicles in public space; now intentionally a low-priority maintenance project. Live: https://ilestoi.pl
- Gesture Inspector — on-device MediaPipe gesture-recognition Android project.
| Area | What I work with |
|---|---|
| Embedded & low-level | ESP32-S3 / C6 / S2, ESP-IDF, PlatformIO, FreeRTOS, BLE, Zigbee, Wi-Fi, MQTT, USB, I²C, SPI, UART |
| Systems & backend | Python, process supervision, Git automation, SQLite, FastAPI, HTTP APIs, WebSocket, Linux/macOS services |
| Android & frontend | Kotlin, Jetpack Compose, TypeScript, React, SvelteKit, Vite, Playwright |
| Applied ML / vision | on-device LLM inference, TinyML, YOLO, OpenCV, MediaPipe, simulation and calibration tooling |
| Hardware | KiCad, SKiDL, PCB automation, sensors, displays, power-management circuits, 3D-printing workflows |
| Verification | host/integration tests, hardware smoke tests, static analysis, sanitizers, coverage, CI and resource diagnostics |
I use AI coding agents extensively, but keep execution and verification explicit. I define constraints and architecture, review changes, debug integration failures and validate behavior with the strongest checks appropriate for each project.
Depending on the repository, that includes host/unit/integration tests, physical-device validation, compiler/static-analysis gates, ASan/UBSan, coverage, memory/resource diagnostics, schema/API validation, reproducible CI builds and end-to-end flows.
Local Agent is part of that workflow: it gives AI-planned work a deterministic, bounded and auditable execution layer across my local repositories.