AI Workflow Reliability
Prompt versioning, lightweight evaluation, and debugging practices for inspectable AI systems.
Computer Engineering · AI Research · Product Building
I am a Computer Engineering student at RWTH Aachen University and a Werkstudent Researcher at Fraunhofer IAIS. I build across prompt workflows, local AI systems, edge deployments, and public-facing desktop tools.
What I am building and researching right now.
Prompt versioning, lightweight evaluation, and debugging practices for inspectable AI systems.
Small, practical utilities for prompt workflows, local experimentation, and reproducible AI tooling.
TinyML and embedded ML pipelines that run outside notebooks on constrained hardware.
Packaged applications such as OpenAnima and MediaRecorder Lite, released for real users.
A quick snapshot of the projects that best represent my current direction.
A lightweight desktop visual companion and local-first animated overlay system.
A real-time visual agent where timing, data collection, and deployment behavior mattered as much as offline model metrics.
Local-first prompt version control with Git-style diffs, labels, metadata, and SQLite.
An end-to-end TinyML audio classifier running on Raspberry Pi Pico and RP2040 hardware.
Small desktop tools packaged and released for real users.
Desktop visual companion
A lightweight desktop visual companion and local-first animated overlay system. It has surpassed 500+ downloads across platforms and is available through GitHub, Microsoft Store, and itch.io.
Vertical screen recorder
A lightweight vertical screen recorder for creators making TikTok, Instagram Reels, and YouTube Shorts content. It focuses on fast capture-area selection, simple recording, and 9:16 export.
The main technical directions behind my projects.
PromptLedger, local LLM tools, reproducible workflows, and debugging utilities for practical AI systems.
Raspberry Pi Pico, TinyML, C++ inference, signal processing, and constrained hardware.
PromptLedger, VAE/VQ-VAE, JumpNet, dataset tooling, and learning through complete builds.
Project write-ups, technical notes, and public learning logs.
Designing a portable prompt ledger with diffs, metadata, labels, and SQLite.
Read write-upFrom dataset preparation and feature extraction to Pico firmware and live inference.
Start seriesCollecting gameplay data, training a vision policy, and running real-time inference.
Start series