Work Experience

Solution Engineer, Cloud & AI: Data Analytics

In my role as Solution Engineer at Microsoft, I work as a technical pre-sales professional. I help customers drive innovation and create business value by designing and validating solutions, leading technical discussion and POCs, and guiding architecture decisions. I am working in the data & analytics area.

Data Engineer, Solution Architect – Unified Data Platform

Company: Vorwerk, A Multinational retail company

Dates: Q1 2024 – Q4 2025

The client is a multinational retail company focusing on household appliances. They operate in more than 60 countries across the world, employ over 9000 people and have an annual revenue of over three billion euro. They also work with approximately 100,000 independent sales partners.

The client has customer, sales partner and employee data spread across many different databases, including SAP systems and an on-premise data center. They want to have a unified single source of truth for this data. They have two existing solutions in different countries, one using an ‘out-of-the-box’ SAP solution that has a costly license and is very limiting for future applications, and another one that requires a total of 48 hours for source data to propagate to the business level. Additionally, they want this consolidated data to be able to source future analytics and AI projects.

I was part of a small team tasked to develop a proof-of-concept within one month. Under a tight deadline, we were able to demonstrate the feasibility and potential of our solution. After the Proof of Concept, I took the lead in designing and developing the data processing with Databricks using PySpark and worked on developing the ingestion pipelines using Azure Data Factory. After a successful deployment in the first market, the same solution is now live in three additional markets, and development is ongoing for three more.

MLOps Engineer – Machine Learning Platform

Company: A Public Transport Company

Dates: Q2 2023 - Q4 2023

The client is a transport company that maintains a public transportation network, which includes buses, trams, metro, and commuter trains. They employ 400 people.

The client wants to use data to determine which are the optimal frequencies for each transit route to service the users in the best way possible. To that end, they want to track the number of passengers on each vehicle to determine whether a route at a point in time is over- or under-utilized, and adapt the offering accordingly. They have collected this data, however; the quality is very low, and there is a lot of missing data. They want to impute the missing data in an intelligent way so this data can be an accurate source for decision making.

I worked as an MLOps Engineer, taking on several responsibilities within the project. I implemented CI/CD pipelines in Azure Pipelines to automate the model deployment process. The model versioning and monitoring in Azure Machine Learning was also taken care of by me, as well as the model packaging in Docker containers stored in Azure Container Registry to ensure consistent deployments. Additionally, I assisted the client's data science team with the data preprocessing and modelling. My input was highly valued as the leading voice on all cloud related matters within the project, as well as strategic insight for future improvements.

Machine Learning Engineer – Emergency Medical Care

Company: A Healthcare Provider

Dates: Q4 2022 – Q1 2023

The client is one of the largest hospital networks in Europe, employing over 26 thousand employees.

The client wants to determine the optimal locations across the city to station their ambulances to minimize the response time. A simulation tool was developed using Azure Functions to test multiple configurations, as well as an ensemble model using scikit-learn to generate the optimal locations.

I worked on the data preprocessing and refactoring of the code, cleaning up the code and documentation. I also rewrote the preprocessing from R to Python, increasing the maintainability of the codebase and speeding up the preprocessing step.

Data Scientist – Time Series Forecasting

Company: A Construction and Development Company

Dates: Q1 2022 – Q4 2022

The client is the biggest construction and development company in Finland. They specialize in apartments, business premises and entire areas, as well as demanding infrastructure projects and paving. Their annual revenue is over 2.2 billion.

The client has several events or objects that lead to a sale, including sales leads and marketing campaigns. They refer to these as ‘opportunities’. Each month, there are sales targets the client wants to reach. The client wants to know, for a certain sales target at a specific point in time, how many opportunities are needed to reach those targets. This way, they can tailor their marketing and work with their salespeople to ensure they are doing what they can to reach the set sales targets.

I was part of a team of two that architected the solution. I was responsible for the data integration and aggregation, cleaning, validating and transforming the data as a source for the machine learning model. I also implemented the feature selection and model training. With two months left to go in the project, the other team member ended up leaving the company, leaving me as the only developer. My contribution was highly valued by the client, as I was still able to deliver the project within the deadline despite the difficult circumstances.

MLOps Engineer – Machine Learning Platform

Company: A Construction and Development Company

Dates: Q1 2022 – Q4 2022

The client is the biggest construction and development company in Finland. They specialize in apartments, business premises and entire areas, as well as demanding infrastructure projects and paving. Their annual revenue is over 2.2 billion.

The client wants to be more data driven and have a central platform for Artificial Intelligence and Machine Learning projects. They would like to be able to re-use the same infrastructure and have a system in place that handles model versioning, packaging, monitoring and deployments.

I developed the Infrastructure as Code using Terraform to deploy the Azure resources, ensuring that it can be easily maintained across different environments and replicated for other projects. Then created the CI/CD pipelines used for model training and deployment. This way, all the steps of the training process, including preprocessing, training and inference, could be automated. Additionally, model versioning and monitoring was automatically integrated with Azure Machine Learning services, and docker images used for packaging and deployments maintained in Azure Container Registry.

Machine Learning Engineer – Computer Vision

Company: A Medical Equipment Manufacturing Company

Dates: Q2 2021 – Q1 2022

The client is a start-up on the cutting edge of AI tooling in the healthcare sector, operating across 17 countries worldwide.

Their flagship product is an AI-powered segmentation tool that helps to standardise contouring and automate it to streamline the radiotherapy treatment planning workflow. Labelling the training data for the model is extremely expensive due to the medical expertise required of the labeler, and the tooling required due to the 3-dimensionality of the data. The project was to leverage large amounts of unlabeled data to improve the model training process.

I worked on an R&D project, in which I created a self-supervised pretraining workflow. This included setting up a preprocessing, pretraining and training pipeline using PyTorch. I tested out multiple existing self-supervised computer vision techniques, as well as altering them to fit the use case of 3D medical data. I also tested different combinations and configurations of proxy tasks, such as masking, blurring, and local pixel shuffling. The final result was a pretraining which was able to improve sample efficiency and training efficiency of the downstream model training. This meant a lower amount of training data and a lower amount of training cycles needed to achieve a similar result.