ML researcher · embodied AI & AI safety

MSc student at ETH Zurich and visiting researcher at Tsinghua, working on vision-language-action (VLA) models and diffusion language models. Pruned up to 90% of VLA visual tokens while retaining success rate. Completed the ARENA curriculum (mechanistic interpretability, LLM evals, alignment science).

education

• 2023 – nowETH ZurichMSc Data Science · GPA 5.63/6 · embodied AI, LLMs, computer vision
• 2025 – 2026Tsinghua University, Beijingstudy abroad · robotics, systems, control, Chinese
• 2020 – 2023ETH ZurichBSc Computer Science · GPA 5.63/6, top 5% · exchange at the University of Washington, Seattle (2022)

experience

• 2026 – nowTsinghua University (THU LeapLab), Beijingvisiting researcher · diffusion language models and VLA models, with Prof. Gao Huang
• 2025Oracle, Zurichresearch assistant · AI coding assistant for the Oracle Multilingual Engine; LLM benchmarks improved 2x with RAG, MCP and agentic loops
• 2024 – 2025think-cell, Berlinmachine learning intern · Attention UNets for a bar chart OCR tool; generative layout models trained on 3M+ slides

projects

• thesisVision token pruning for VLA modelssubmitted · prunes up to 90% of visual tokens and retains the success rate in simulation and real-world tests; beats the state of the art in FLOPs and latency
• courseARENA 3.0, Alignment Research Engineer Acceleratorall chapters: transformer interpretability, LLM evaluations, alignment science, RL
• 1st placeETH Datathon Kaggle competitionYOLO/ResNet pipeline extracting pie chart percentages, within 24 hours
• thesisLLM-augmented contrastive learning5M+ paraphrases generated with LLMs; fine-tuned BERT models beat the benchmarks on semantic similarity

skills