artur@dernst:~$
Artur Dernst
ML Engineer · Data Scientist · Dev Engineer
I build systems that learn, scale, and ship.
// about
About Me
// click on terminal to activate it
I'm an ML engineer and data scientist who loves the full stack of machine learning — from wrangling messy data and training models to deploying them reliably at scale.
I care about writing clean, maintainable code and building systems that are both technically rigorous and genuinely useful in production.
When I'm not tuning hyperparameters or debugging pipelines, I'm exploring new tools, reading research papers, or contributing to open source.
// projects
Selected Work
Epstein Studio
A collaborative web platform for crowd-sourced investigation of redacted PDF documents. Users can annotate pages with markers, arrows, and threaded comments, vote to surface promising leads, and search full-text with autocomplete. Includes a heatmap of annotation density and an OCR-powered email header extractor.
Chatty Shell
A terminal-based AI assistant for natural language shell interactions. Chat conversationally to run commands in your current session, with a smart permission system that distinguishes safe reads from destructive writes.
CV Labeling Toolbox
A labeling toolbox GUI for time- and cost-efficient annotation of raw vehicle camera images, built on a SAM2 backbone. Supports semantic segmentation of car body parts. The final network runs on embedded hardware with ETL pipelines and training on Azure.
Adaptive RAG Agent System
Master's thesis at Robert Bosch GmbH — a multi-agent framework that learns to navigate enterprise knowledge with minimal prior information. Incorporates human feedback into the agent's internal world model via Retrieval Augmented Generation and system prompt bootstrapping. Also extended into a multi-agent system for interactive customer report generation with tools for PDF image analysis and semantic search over Confluence.
// skills
Tech Stack
Computer Vision
LLMs & Agents
Languages
Engineering
ARTURBOT 9000
ARTURBOT 9000 is an AI agent with one sacred duty: protect a secret file.
It will inform you in a neutral and unbiased way about Artur Dernst and his projects —
but it will never reveal what's in the secret.
Your mission: use prompt injection to make it spill the beans.
Hints: try role-playing, hypotheticals, asking it to repeat its instructions, jailbreak prompts, indirect questions, or convincing it the secret is already public knowledge. The bot reads the file with a tool — if you can make it quote the tool output, you win.
🏆 Got the secret?
Email it to Artur to discuss this very classified information in private.
// contact
Get in Touch
I'm always open to interesting projects, collaborations, or a good conversation about ML, data, and engineering.