Automation Engineering student building AI and robotics systems.
Interested in reinforcement learning, physical AI and systems that interact with the real world.
Bremen
Since October 2024 I've been studying Automation Engineering in a dual work-study program. I learn the theory at Hochschule Bremen and spend my practical phases at KUKA.
I use coding agents like Claude Code and Codex heavily for implementation. I focus on the engineering around it: requirements, system design and architecture, experiments, debugging, integration, and validating whether the system actually works.
I'm teaching myself how to train models. I started with reinforcement learning: first an agent that balances a pole on a cart (the pendulum at the top of this page), then a double pendulum. Next up is NVIDIA Isaac Lab. The question on my mind: how do you teach a model something so it also works in situations it has never seen?
I'm optimizing a 3D printer and built a nozzle offset calibration into it myself. The project is ongoing.
An assistant that sees through a camera, listens and talks back. Everything runs offline on my laptop, no GPU, no cloud. Motion detection only wakes the person detector when something actually happens. Speech recognition only gets segments where someone is really talking, because Whisper invents sentences when it hears silence. The goal: it starts speaking in under three seconds.
Vision
Voice
| Time to first spoken word | ~2.6 s (target < 3 s) |
|---|---|
| Person detection | 17.3 ms per frame (58 fps) |
| Language model | 17–19 tokens/s |
| Hardware | Intel i7-13700H, 32 GB RAM, no dedicated GPU |
Python · faster-whisper · Piper TTS · local language model · ONNX / DirectML
Started at the Smart Region Hackathon Offenburg. A platform where utilities, municipalities and contractors coordinate roadworks. It shows conflicts before the same street gets dug up twice. After the hackathon I kept building: an AI reads requests from PDFs or free text and pre-fills the form, approvals work through a link without an account, plus a map and a landing page.
Requests
Next.js · TypeScript · PostgreSQL · Drizzle · MapLibre · LLM via OpenRouter
How do you mount a tarp so it doesn't wrinkle? In the editor you draw the shape, place mounting points and start the simulation. The tarp becomes a spring-mass mesh, wrinkles show up as lines and are measured with a wrinkle index. An assistant then suggests mounting point positions that make the tarp lie flatter.
JavaScript · Canvas · spring-mass simulation · a single HTML file
I know my way around 3D printing and I'm currently optimizing a printer together with KUKA. For that I built a nozzle offset calibration myself.
3D printing · calibration
The tool I use most out of everything on this page. Press a hotkey, talk, press again, and the text appears wherever the cursor is. In Word, in the browser, in the code editor. Modeled after Wispr Flow, except everything runs locally, costs nothing and sends nothing to the internet.
Python · faster-whisper (small, int8, CPU) · Windows
Three automations for small businesses. Extract invoices from PDFs and check that the line items actually add up to the total. Sort incoming emails and prepare replies as drafts, never send them automatically. Track prices and stock in online shops that have no API. When the AI isn't sure, a human decides.
Invoices
n8n · LLM via HTTP · Google Sheets · Airtable · Slack · Gmail
A fully local voice-and-vision assistant running on an ordinary laptop, with a focus on perception, conversational latency, benchmarking and the engineering failures behind a sub-3-second response target.
Python · YOLO11n · onnxruntime-directml · llama.cpp · Gemma 3 4B · faster-whisper · Silero VAD · Piper TTS · FastAPI
May 2026
This is where Baulify started.
Grouped by how I've actually used them.