Copenhagen, Denmark | DTU M.Sc. Autonomous Systems

I build robotics, sensing, and data systems that hold up outside ideal conditions.

My work sits where software, physical systems, and messy data meet. I care about evaluation, simulation, useful interfaces, and making technical logic legible to the people who need to use it.

View selected work About Resume

Robotics softwareSensing + data qualitySimulation + evaluation

Selected work

Four projects worth starting with.

They show the mix of product thinking, research discipline, and field-facing engineering most clearly.

Consumer product + environmental modeling

SunnySips turns solar geometry and weather into a product people can use in seconds.

Built around sun position, urban occlusion, and weather attenuation, then shaped into a consumer interface that feels fast and obvious instead of technical for its own sake.

Modeled live outdoor conditions rather than relying on static cafe lists.
Tightened defaults, contracts, and snapshots to make the experience feel trustworthy.

SwiftUIFastAPIGeospatial logicWeather systems

Product direction + systems thinking

TRYBE explores how product, brand, and system design can lower the friction of trying new things.

The interesting part is the coherence. Product behavior, mobile structure, backend planning, and visual language were treated as one system with real constraints around adoption, safety, and repeat use.

Designed sessions, perks, and partner logic as one believable loop.
Used UX decisions to make the concept feel concrete instead of aspirational.

Product directioniOS architectureBackend planningDesign system

Thesis object

Soft-finger actuator model

Simplified modeling for rehabilitation-oriented design choices.

Define assumptions

Sweep parameters in simulation

Compare against bench measurements

Model assumptions

Simulation sweeps

Bench checks

Research + modeling

A simulation-first thesis on soft-finger actuation for rehabilitation-oriented hardware.

Current DTU thesis work focused on simplified, interpretable modeling before physical iteration gets expensive. The emphasis is on explicit assumptions, bounded claims, and bench checks that keep the model honest.

Using simulation to narrow tendon routing and stiffness decisions.
Treating bench validation as a calibration step, not decoration.

DTUBiomechanical modelingSimulation-firstBench validation

Open thesis PDF

Robotics + data quality

PerPlant combined field sensing infrastructure with cleaner computer-vision dataset design.

Two distinct threads mattered here: ROS2 thermal capture with GPS metadata in the field, and a separate curation workflow for choosing smaller, more representative annotation sets from highly redundant agricultural imagery.

Thermal + GPS collection was built around real field conditions, not ideal capture.
Filtering, embeddings, clustering, and grouped splits were used to reduce leakage and redundancy.

ROS2 sensingThermal + GPSEmbeddings + clusteringLeakage control

Collection

ROS2 thermal imaging

Thermal imagery published through ROS2 in C++ and Python, with GPS geotagging folded directly into the field workflow.

Curation

Representative subsets

ROI filtering, detector embeddings, UMAP, HDBSCAN, and grouped splits used to shrink annotation batches without losing coverage.

Away from work

Travel, water, mountains, movement, and cooking keep the rest of the work grounded.

Outside work, I spend a lot of time around places, routines, and hobbies that keep curiosity concrete.

Open life page Travel atlas

Contact

Open to robotics software, sensing systems, simulation, and applied research work.

About s242507@dtu.dk