Sebastian Hedegaard Hansen

Resume

Here you can either download my resume or read a more detailed version

Solution Architect Zevit. (Aug 2022 – Current)

ALeading the design and development of the software Asset Integrity Hub (AIH). Using AIH, Vestas achieved a saving equivalent to $28,000 per wind farm annually in their service business. ZFs sparepart service reduced the time allocated to data engineering by 75% (3 full time employees). We did this by building a user focused software that reduced the time spent on failure management, maintenance execution and analytics. Enhanced organizational data integrity was realized by aligning the software’s data models with industry standards like ISO/IEC 81346-10:2022.
Managed a team of 25 IT professionals in Armenia instilling a culture of ownership throughout the product lifecycle to minimize handoff cost and to encourage the sense of responsibility.
Engineered a series of operational solution for our customer. For example, a high-capacity event-based integration between AIH and SAP, handling over 5 million requests daily.
Bridging the gap between the end-user’s and the technical product teams I initiated new product functionality by using my domain knowledge and technical abilities to design, develop and validate prototypes minimizing the time from idea to finished feature.

Solution Owner Zevit. (AUG 2021 – Aug 2022)

Led a central project for Vattenfall that materialized into a knowledge register used by over 1000 engineers. The register was built to create more intelligent maintenance plans that utilizes the operational data coming from technicians, SCADA systems and asset failures.
Designed and implemented a robust microservice architecture, enabling the seamless deployment of analytical models and real-time asset monitoring, accessible to both technical and non-technical users.

Student helper ZEVIT. (Feb 2021 – Aug 2021)

Working as a cloud developer at ZEVIT while writing my master thesis:

Working with bigger international client like Vattenfall and Semco setting up infrastructure in azure to run deep learning models to detect corrosion on off shore substations
Gaining knowledge about industry specific data models like RDS and IEC to guide ZEVIT’s customers in their data management.
Responsible for the daily operation of ZEVIT's software Asset Integrity Hub at Vattenfall, responding to feature request, bug reports etc.

Consultant Nerve Smart Systems. (Feb 2018 – Aug 2018)

To gain some practical experience working with bigger technology projects a fellow student and I worked with the newly founded company Nerve smart system:

I consulted on the overall design and functionality of a e-mobility charging station.
I planned and executed 2 different user-centered design processes. Working with Adobe XD to get the design right before moving to Xamarin for the cross-platform implementation.
Overall, I got to learn what it means to plan and execute a larger development project that needs to be finished at a certain deadline.

Student assistant Pfizer Denmark. (Sep 2016 – Dec 2017)

At Pfizer I assisted my colleagues in the sales department with analyzes of relevant sales data and helped prepare the different sales material.

I developed a series of programs to automate the task of collecting, cleaning and analyzing larger amount of sales data In python, allowing me to constantly devote my time to new tasks.
I worked with a group of people from India in maintaining and updating a salesforce CRM system (Veeva).
I got experience with the basic terminology and tools within sales, marketing and budgeting.
I provided support to my colleagues in their daily work by sharing the different optimizations techniques I acquired at school.

Assisted at COWI. (Sep 2014 – Aug 2015)

Newly graduated from high school I worked full time at the engineering company COWI as an assistant in facility management.

I developed a series of systems to optimize my workday, this include writing a program that automatically alerted me when a printer was running out of toners, instead of me having to check manually.
At my first real job in a large corporation, I learned how company culture influences the overall work environment and how I should you navigate in the workplace.

Badminton trainer “Birkerød Badminton club”. (2008-2014)

Planning and executing the daily training for a bigger group of kids and teenagers.

Paperboy. (2007-2008)

Got the discipline to wake up early every Saturday morning.

Dancer at the Royal Danish Ballet School (2005-2006)

Daily training with focus on discipline and physical skills.

Master student at the Technical University of Denmark – Human-centered artificial intelligence. (2019-2021)

Acquiring practical knowledge about the technologies related to digital media engineering and the related business aspects to be able to develop new ways for people to interact with technology through software.

Advanced the precision of predictive ML algorithms by developing synthetic data augmentation techniques using generative AI, which improved model accuracy by 2% on benchmark data sets. Recognized for contributions to the field by industry peers at NVIDIA.
Analysis of large connected networks of Reddit comments using network science and natural language processing. Visualized extreme complex datasets using standard plotting tools, word clouds and graphs.
Using multiple CPU’s and GPU’s to compute large calculations, optimizing code and compilers to run more efficient by using openMP and CUDA.
Theoretical design of algorithms for analyzing and storing large datasets.
Using modern machine learning tools to analyze data and making work processes more efficient or automated.
Design and implementation of larger software systems using UML and a range of functional and object-oriented programming languages.

Bachelor student at The Technical University of Denmark – Design and innovation. (2016-2019)

Aspire to create the products of tomorrow through creative processes in close contact with the end-users. Use the technical skills we acquire in class to create innovative products both physical and digital.

Strong technical foundation within math, mechanics, material science, physics and computer science.
Focus on the planning and execution of larger projects in close collaboration with end-users.
Developing a smart connected (MQTT technologies) baby monitor to monitor sleep patterns of new parents with the purpose of only waking the least affected by a crying baby.
Thesis: planning, developing and executing an extensive between subject experiment to gather data about people’s ability to drive a gaze-controlled telepresence robot. Functions was implemented and data was collected using Unity 3D written in C#.

Exchange student New Jersey Institute of Technology. (Aug – Dec 2018)

Lived on Manhattan and went to school in New Jersey for a semester.

Living in one of the largest melting pots of human potential in the world, meeting new people with very different backgrounds than my own, allowing me to become more openminded and curious.
Developing a large software system that grades a university students python-functions like “code judge” or “Hakerrank”. The functions were evaluated using regular expressions and data was sent between the different servers and databases using “cURL” commands all written in PHP.

Bachelor student at Copenhagen business school - Organizational communication. (2015-2016)

Strategically working with communication in organizations to create connection between their overall goals and the way they communicate.

STX at Birkerød Gymnasium (2011-2014)

Python:

The first language I got acquainted with was python. I took a summer course (see “Introduction to programming and data processing” under courses below) where I learned the fundamentals of programming and got a sense of the mindset you need to have as a programmer. I quickly figured out that programming was a very creative discipline framed by some very clear boundaries.

I used the skills acquired in this class at my work at Pfizer where I wrote programs that collected data from our wholesaler’s online platform using the library “beautiful Soup”. I cleaned, organized and analyzed the data using “pandas”, “math” and “NumPy”.

Since then I used Python for a variety of courses. For example in the course “Social graphs and interactions” where I used the “Networkx” library to analyze and visualize graphs.

I have used python extensively for building deep learning models in pytorch in my thesis and deep learning courses.

Java, C and C#:

I took a Java and C programming course where I learned about classes, objects and methods and how they could be used to become a more efficient programmer.

In relation to my bachelor thesis I used C# to implement an experiment in Unity 3D. The test subjects where to wear a VR headset with build-in eye trackers. We had to create objects within the VR environment that the users could interact with. For this we used some of the features in Unity and modified them with a C# class. Along with this we wrote a program that collected all the relevant data about the test subject’s behavior – reaction time, movement, gaze position etc.

At ZEVIT I have used C# to implement a REST API for creating and maintaining analytical models and issuing monitors based on the models.

Arduino (based on C++):

I took a course in mechatronics where we learned about microcontrollers, actuators, sensors and IoT technologies. In a group we created a series of smart products, a speaker, a locker and a baby monitor. As the most experienced programmer I was primarily responsible for creating the code, both for the microcontrollers and the web-based logic. Here I used Node-red to handle the incoming and outgoing MQTT messages from the different microcontrollers.

I thought this was so fun and cool that I continued to develop small projects that where useful in my everyday life. Recently, I merged my passion for cooking with technology by building an automated indoor greenhouse, termed a 'food computer, That also functions as a fermentation chamber for my koji, misos and kimchi.

PHP:

During my semester in New York I took a course in software engineering where we had to create a system written in PHP. A professor should be able to log in, create a programming question for an exam and create the exam from the available questions. A student then had to log in and take the exam. The system would then run the program on a series of testcases and check for the most common errors (printing instead of returning etc.). I was responsible for writing the “middle-end” – part grading the students answers and parsing information between the “front-end” and “back-end”.

I took a course in high performance computing where we wrote all our programs In C due to its possibility to write extremely efficient code. We used openMP to write programs that ran on multiple CPU cores and later using CUDA to write programs that ran on multiple cores on the GPU.

SQL:

I took a course in database design where I learned about the fundamentals of relational databases. Here I learned to create ER-diagrams and implement them using SQL. I have later used some SQL, to amongst other thing, create a login page for one of my code-practice projects.

Listed below are the most relevant courses taken on my bachelor and master.

Algorithms for Massive Data Sets

System Integration

Introduction to Machine Learning and Data Mining

Social graphs and interactions

UX Design Prototyping

Commercialization of high tech concepts, entrepreneurship and science in action

High-Performance Computing

Participatory Systems Design

Mechatronics Engineering Design

Introductory Programming

Algorithms and Data Structures

Advanced Engineering Mathematics

Advanced Engineering Mathematics 2

Introduction to programming and data processing

User-oriented design

Deep Learning in Computer Vision

Computational Tools for Data Science

Deep learning

Advanced Project in Digital Media Engineering

Data analysis of IoT sensors in building complexes

Outside my professional life I have a series of interest that I believe captures my personality. For many years I have spent the majority of my spare time in the kitchen experimenting with fermentation and different styles of cooking. I love to source my own ingredients by going spearfishing or collecting herbs and vegetable from my parent’s garden. With all this food in my life I try to stay reasonable active in my everyday life. Most reasonably I caught the crossFit fever.

My curiosity and motivation to experiment are not only expressed in the kitchen. I often get interested in various subjects ranging from 3D printing to the American political system. I spend a good amount of time watching videos/documentaries and reading about different subjects that don’t have anything to do with my field of study. This is especially an advantage when I play “bezzerwizzer” but also (arguably less important) during creative processes where interesting connections can be drawn from different subjects.

This diverse range of interest have resulted in a network of friends with very different backgrounds than my own. I love meeting these friends over a cup of coffee or a beer discussing new perspective on different subjects and events.

Download Resume

Bachelor thesis: Gaze controlled telepresence robot

The use of gaze to control a telepresence robot could provide motor disabled people with a new level of mobility. With a well developed robotic telepresence system that uses gaze as a control input, individuals with motor disabilities could be able to occupy jobs operating the robots in various situations. This raises the question if virtual training environments could be used to educate this whole new kind of workforce. This report investigates if simulated envi- ronments potentially could be used to replace training in reality when educat- ing robot operates. In order to evaluate operators performance, alertness and awareness new measures needs to be developed. A primary focus in this report will therefore be to investigate what information that can be extracted from implementing a pro-saccade test in the same HMD that the operator uses to control the robot.
To investigate these aspects a between-subjects experiment was conducted with 32 participants operating a gaze controlled telepresence robot. The experiment showed that the operators task completion time was independent of the environment they had been trained in, only the number of collisions was higher for the test subjects trained in VR.
The results of the pro-saccade test remained constant between trials for each test subject, and thus no learning effect was seen. But the results of the pro-saccade test indicated that the best robot performers had shorter latencies on their saccades indicating higher alertness and was experiencing lower workload than the worst robot performers. The experiment showed interesting results but further research into the correlation between eye movements and performance, alertness and awareness needs to be done. Based on the finding from the experiment 5 possible concepts was developed, 3 concepts that includes the use of a dynamic saccade test, 1 concept of a VR training environ- ments and 1 concepts of a interface that uses visual stimuli to prim users to make saccades. With further testing, these could be implemented in the robots operating system to enhance the user experience of driving a gaze controlled telepresence robot.

Data-Driven Failure Diagnostics of Industrial Assets

When industrial assets malfunction, the financial consequences can be immense as operations poten- tially need to temporarily shut down or slow down. With assets becoming more complex the need for digital troubleshooting systems become more apparent as humans are limited by their weak ability to comprehend and analyse a large amount of complex data. Such a system needs to detect when a failure has occurred, diagnose the most likely failure, guide a technician or operator in resolving the failure, and lastly test if the asset is operating as normal after the repair. This thesis focuses on the diagnosis of the assets based on the multivariate time series data coming from the asset’s sensors. An extensive set of experiments were carried out on both open-source datasets and a real-world dataset from five wind turbines. An automatic prepossessing and hyperparameter tuning process of multiple state-of-the-art time series classification algorithms, including InceptionTime and MiniROCKET, are presented. The method is proven to be effective on the open-source dataset where the achieved results match those from published reviews of the same algorithms using the same datasets. The method was used on the real-world data where it was shown that it was possible for the classification algorithms to identify, to a wide extent, in which component of the wind turbine the failure had occurred. Failures on these industrial assets can be rare. The classification models will therefore have too few examples to learn the general characteristics of a failure. Therefore, the use of generative adversarial networks (GANs) to increase the performance of the classification models was investigated. For this, a version of the newly published generative model, uTSGAN, was developed and compared to the original. The experiment showed that it was possible to train a GAN that was able to generate realistic samples that could improve the performance of the classification model when combining it with the true data. More work however needs to be done in the academic community to make it feasible to use GANs in an operational setting.