Resume

Lukas Müller

Frankfurt am Main, Germanycontact@lukasmueller.devGitHub

Education

Bachelor of Science in Computer Science
Technical University of DarmstadtDarmstadt, HE
  • Current grade: 1.1 German scale (GPA: 3.9)
  • Upcoming thesis at the Intelligent Autonomous Systems Group (Prof. Jan Peters), starting Aug. 2026; research area: vision-language-action models for humanoid loco-manipulation
  • Exchange semester: Aarhus University, Denmark, Jan. 2026 – Jun. 2026
  • Elective courses: Robot Learning, Visual Computing, Markov Decision Theory, Machine Learning in Game Theory
Bachelor of Arts in Digital Business Management
Baden-Wuerttemberg Cooperative State UniversityMannheim, BW
  • Final grade: 1.2 German scale (GPA: 3.8)
  • Thesis: Conception of a Strategy to Increase Data Quality for a Predictive Analytics Application

Work Experience

Working Student Machine Learning & Data Engineering
Compredict GmbHDarmstadt, HE
  • Implemented a contrastive timestamp-embedding model in Keras for a production time-series package
  • Built a testing harness for evaluating the performance of a data quality platform
  • Developed Python packages for data quality and data labeling for high-frequency time-series data, including core logic, tests, and documentation
Working Student Bioinformatics Research & Development
BioNTech SEMainz, RP
  • Developed a data pipeline transferring local research files to AWS and making them accessible in a data warehouse
  • Extended a bioinformatics feature pipeline used for downstream machine learning workflows
Cooperative Education Student
Fresenius GroupBad Homburg, HE
  • Completed internship rotations in Project Management, Business Intelligence, and Process Mining

Projects

agentic-dev-toolkit
Open Source
  • Built vibe, a CLI giving each task its own git branch, worktree, and tmux session, with handoff between machines
  • Authored a library of portable Agent Skills for the open Agent Skills format, covering e.g. codebase health checks
  • Technologies: Bash, tmux, git, shellcheck/shfmt, bats
mini-vla: A Tiny, Language-Conditioned Vision-Language-Action Policy
Open Source
  • Designed a language-conditioned spatial-attention policy that fuses a rendered scene, a typed command, and gripper state to control a 2-link arm on block-grasping tasks
  • Trained the policy via behavioral cloning against a closed-form analytical inverse-kinematics expert, then ported the identical architecture to TensorFlow.js so it trains live in-browser on my portfolio site
  • Technologies: Python, Keras, TensorFlow.js, Weights & Biases
Large-Scale Robot Generation (Team Project)
TU Darmstadt (Intelligent Autonomous Systems Group)
  • Built an MPNN-based variational autoencoder learning latent representations of robot morphologies in MuJoCo
  • Implemented training and evaluation workflows in PyTorch with experiment tracking in Weights & Biases
  • Technologies: Python, PyTorch, MuJoCo, Weights & Biases

Skills

Machine and Robot LearningPyTorch, TensorFlow/Keras, MuJoCo, Weights & Biases
Dev Toolinggit, CI/CD (GitHub Actions/GitLab CI), Docker, Claude Code
Programming & Data EngineeringPython, C++, SQL, AWS (S3/Glue/Redshift/MWAA)
LanguagesGerman (native), English (fluent), Spanish (basic)