About Me
I'm a dynamic and versatile Machine Learning Engineer with extensive expertise in machine learning applications, infrastructure, applied machine learning research, data science, and software engineering. I have experience in architecting distributed, LLM-driven systems, large-scale model training, and data processing pipelines. On top of my professional and practical experience, I bring a strong academic foundation in Machine Learning and Data Science from the University of Copenhagen.
I have done academic research on mechanistic interpretability of Transformer language model dynamics, and I have done personal projects ranging from compiler design to voice synthesis to evolving artificial organisms driven by neural-networks. Among my peers, I'm known for my optimistic, persistent, and proactive approach and my work ethic, along with my ability to quickly learn and apply new concepts and technologies to solve complex and exciting problems.
I'm motivated by real impact and aspire to become the wisest and most knowledgeable engineer I can be.
Experience
Ogment AI
Senior Machine Learning Engineer | Zurich, Switzerland | Aug. 2024 - Present
- Member of the founding engineering team and architect of the company's core product.
- Responsible for designing and implementing a distributed, LLM-powered, actor-model, multi-agent collaboration framework for reliable and interactive workflow execution.
- Developed and executed a billion-scale applicant profile processing and ingestion pipeline, efficiently indexing the world’s working population using rich semantic embeddings and automatically derived smart attribute tagging. This enabled agentic LLM-driven agents to conduct candidate searches using structured natural language queries.
- Designed and implemented a scalable, serverless architecture for multi-agent workflow processing, enabling real-time frontend interfaces, utilizing Google Cloud Pub/Sub, Cloud Run, Cloud Scheduler, Docker, and MongoDB composed with Terraform and integrating with a FastAPI API.
Unity
Senior Machine Learning Engineer | Zurich, Switzerland | Sep. 2023 - Aug. 2024
- Drove development of the Muse LLM platform, responsible for scaling Muse Chat and onboarding 14 developers, launching alpha, beta and generally available versions of Muse Chat - enabling and accelerating creation with natural language in the Unity Editor.
- Designed and implemented a universal multi-modal data ingestion and perception system enabling responses with text and images, leading efforts to extend and scale ingestion pipelines and lowering the barrier of entry for new internal engine documentation to be included in Muse Chat.
- Developed distributed model training tooling and infrastructure compatible with Azure, enabling fine-tuning of specialized language models and conducting experiments with model distillation and model merging.
Unity
Machine Learning Engineer | Zurich, Switzerland | Sep. 2022 - Sep. 2023
- As the founding engineer of the Muse Chat team, architected and implemented the Muse Chat backend, including the design of its unique contextual retrieval-augmented generation system; developed data ingestion pipelines for all Unity core and package documentation repositories.
- Developed an emotional speech synthesis pipeline for privacy-preserving speech and emotion recognition evaluation data and built acoustic toxicity detection systems as part of the Safe Voice ML team.
Unity
Software Engineer | Zurich, Switzerland | Mar. 2022 - Sep. 2022
- Designed, implemented, and evaluated federated machine learning training pipelines for automatic speech recognition models, resulting in extensions to Unity's proprietary sequence modeling framework used in production.
- Developed simulation tools for modeling online gaming communities and social networks with live data visualization and analysis, leading to a proof-of-concept product that informed research in multiplayer services.
Valuer AI
(part-time) Data Scientist | Copenhagen, Denmark | Sep. 2021 - Feb. 2022
- Responsible for implementing automated workflows for semantic data labeling of hundreds of thousands of company profiles.
- Developed a service for hierarchical clustering and semantic record-linking, and company characteristics to track potential duplicate data entries.
Kvalifik
(part-time) Full-Stack Developer | Copenhagen, Denmark | Jul. 2018 - Jul. 2020
- Responsible for development of version 2.0 of Kvalifik's freelance gig economy platform using React/Gatsby with TypeScript.
- Worked on building and maintaining the backend of Kvalifik's gig economy platform using Django, working with PostreSQ ORM.
- Built high-performance automated text processing tools using Rust extracting elements for internationalization across multiple codebases.
- Communicated and aligned with design and product teams in a very dynamic startup environment.
- Helped build the initial version of Kvalifik's freelance gig economy platform using Django and Vue.js.
Personal Projects
Wu
Developed a compiler and type-checker for Wu, a Rust-like programming language tailored for game development. The project has gained over 459 stars on GitHub.
feature-dynamics
Research on mechanistic interpretability of language model feature adaptation and merging dynamics, offering insights into the workings of Transformer models.
speech-to-code
Production-grade speech-to-code service, enabling users to generate code through voice commands.
clsp
Experimental contrastive language voice and emotion pre-training for voice diffusion models.
Orto
Grammar correction and punctuation correction service for Danish, using custom RNN model for efficient punctuation restoration on part-of-speech tokens, automatic dependency tree analysis, BERT-based word substitution, and more. Developed during my final year in high-school and successfully tested on graded Danish essay assignments.
Screenshot of Orto frontpage and service in action.
... and many more on my GitHub.
Publications
Research paper on mechanistic interpretability of language model feature dynamics and merging dynamics, offering insights into the workings of Transformer language models.
High-level technical blog post on the Muse Chat LLM integration and contextual retrieval-augmented generation system, published on the Unity Blog.
Note: I drafted the technical paper and initial draft of this blog post, and I used to be cited as a co-author. Today it no longer shows my name on it.
Education
University of Copenhagen
B.Sc. in Machine Learning and Data Science
Thesis Grade: ECTS A (View Paper)
Sep. 2021 - Jul. 2025
Copenhagen Business College
Business and Economics
GPA: 11.2 of 12 (top 1%)
Sep. 2018 - Jul. 2021
Skills
ⓘ You can click on skills to learn more.
Programming Languages
Programming since age 10, professionally since 2018, with experience in multiple languages.
Machine Learning
Professional experience in applied ML, including large-scale distributed training with PyTorch, federated learning with TensorFlow, and large scale data processing with scikit-learn.
Data
Experience with large-scale data processing and analysis, and reliable and compliant data management.
Backend
Experience with productio backend development, including REST APIs and distributed web services.
Infrastructure / DevOps
Experience with machine learning- and general cloud infrastructure and DevOps tooling, including infrastructure as code.
Languages
🇩🇰 Danish (native)
🌎 English (bilingual fluency)
🇫🇷 French (beginner)
🇨🇭 German (beginner)
Interests
- Brazilian Jiu-Jitsu (50+ competition fights)
- Weightlifting
- Hiking with my dog
- Pizza baking