Switching Board
A USB- and Ethernet-controlled laboratory instrument built on STM32H563 and Zephyr.
- 20-terminal relay matrix; 1-, 2- and 4-wire modes
- 14 input / open-drain digital I/O lines
- Browser UI, HTTP API, SCPI and HiSLIP
ANALOG VECTOR / ENGINEERING LAB
Hardware, speech, and video. A relay switching instrument, private speech recognition, and a native video decoder. Three projects connecting the physical and digital worlds.
Native software. Connected hardware. Real engineering.
// 01 — PROJECTS
From laboratory automation to local AI and media playback. Explore the capabilities behind each project.
A USB- and Ethernet-controlled laboratory instrument built on STM32H563 and Zephyr.
STT STUDIO / WINDOWS
Native speech-to-text for Windows and Android. Your audio stays on your device.
A native H.266 / VVC Main10 decoder integrated with Windows Media Foundation.
Repository-backed feature summaries. Engineering diagrams are illustrative; the STT image is an actual application screenshot.
// 02 — OUR APPROACH
Different technologies, the same engineering discipline: understand the problem, design with intent, and validate with evidence.
Define the problem, the inputs, and the constraints that actually matter.
Bring architecture, electronics, and algorithms into one coherent solution.
Turn ideas into a prototype and make its components work together.
Compare results with the goal. Iterate on what the evidence shows.
HARDWARE × SPEECH × VIDEO
PRJ—001 / EMBEDDED INSTRUMENTATION
Signal Dynamix / STM32H563 + Zephyr
A laboratory switching instrument with a 20-terminal relay matrix, fourteen digital I/O lines, measurements, and a common control core shared across USB and Ethernet transports.
Engineering prototype, not a production-certified USB, LXI, EMC, or safety product. Electrical ratings and calibration still require hardware validation. Unauthenticated Ethernet control is intended for trusted instrumentation LANs.
PRJ—002 / SPEECH & LOCAL AI
Private speech-to-text / Windows + Android
Native applications for local speech recognition with Whisper Large v3 Turbo Q8_0 and Canary 1B v2 INT8. Inference runs on CPU; the desktop application does not require Python or FFmpeg.

The screenshot shows Windows. Android also provides local Whisper / Canary inference, waveform seeking, recording, editing, and export; audio formats depend on device decoders. Speaker diarization is a Windows feature. Model downloads require internet access.
PRJ—003 / NATIVE VIDEO DECODING
H.266 / VVC Main10 · Windows Media Foundation
An x64 Media Foundation Transform backed by VVdeC, bringing VVC Main10 decoding into the Windows media pipeline and modern Media Player.
This project is a decoder, not an encoder. The production codec uses VVdeC; the independent scratch_codec research decoder is not linked into the Media Foundation codec. Playback requires installing and registering the Windows codec package.