Henrik Svensson

Developed a service for semi-automatic column mapping in CSV files from POS systems. The service was designed to handle multiple types of CSV files, including sales, customers, and items. The AI model used a combination of deep learning, feature engineering, and lookup tables to accurately identify each column's content.

I also created a simple front end using Vue to allow users to interact with the service. I containerized the project using Docker and deployed it to the Google Cloud Platform. In addition, I set up a database to store the lookup table, collected data to build the lookup tables, and used a character-based long short-term memory to predict unknown values such as brand names and first names.

As the lead developer on this project, I implemented the AI model, designed the front end, and managed the deployment and database setup. This project required a strong understanding of machine learning and web development technologies.

Created a sales and inventory forecasting system using various machine learning models. The product incorporated a simple SARIMA model and other deep learning models. I used Flask-RESTX to build a web API for the project, which was then deployed to the Google Cloud Platform using Kubernetes.

As the lead developer on this project, I led a team of students and implemented and evaluated various machine learning models. This project required a strong understanding of data analysis and machine learning techniques, including Python and TensorFlow.

Architected and developed an internal ranking system for customer support tickets that extracts tickets and emails from a CRM system. It uses NLP techniques to extract key content and prioritize customers based on the urgency and emotions expressed in their requests.

The system ranks tickets and alerts the support team to the most pressing and emotionally charged cases. As the lead developer on this project, I handled the design and implementation of the service and integrated it into the CRM system. This project required strong skills in NLP, web development, and data management.

Joined the company and discovered a lack of structure that led to issues such as direct pushes to master without clear commit messages, conflicting pushes, and different coding styles. To address these issues, I proposed and implemented a new workflow for the team. This included using feature branches to ensure that code changes were adequately reviewed and tested before merging into the main branch.

I also introduced code reviews to help ensure that code changes are high-quality and compliant with the coding style. To achieve consistency in coding style, I suggested using Black, a formatter that ensures all code adheres to PEP8, the official Python coding style guide.

To further improve collaboration within the team, I introduced Scrum, which kept the team on track and focused on their goals. I also spent quite a lot of time on code reviews, providing constructive feedback and tips on improving their code. In this way, I helped improve the development team's overall quality and productivity and ensured that the codebase was clean and maintainable. In addition, I set up a PyPI server that enabled the team to use reusable libraries.

Handled the architecture and development of an automatic end-to-end migration service between various POS systems. The service was designed using a microservices architecture, with separate services for monitoring and data fetching, conversion, and uploading. I developed the data fetching and conversion services, which fetched data related to customers, items, and sales and converted it to our standard format.

I also conducted code reviews and mentored junior team members who developed the data uploading and monitoring services and the API service for communication with the front end. As the primary developer on this project, I played a crucial role in guiding the service development and ensuring its success. This project required a strong understanding of microservices architecture and data management.

Contributed to an integrated POS system that allowed local businesses to easily incorporate an online store for curbside pickup orders into their current POS system. The system allowed customers to order items online and pick them up without having face-to-face interactions.

This project was developed in response to the COVID-19 pandemic to make shopping safer and more convenient. The POS system was designed to be easy to use for both customers and business owners and required a strong understanding of web development and data management.

Built a system to detect fraudulent physical manipulations on documents, specifically punched holes. To maintain a low false positive rate of 0.001%, I collected data and trained and evaluated multiple models using deep neural networks and the Hough transform. I implemented a map to reduce false positives by assigning lower probabilities to common false positive locations.

The first version of the system, which was brought to production, achieved a high true positive rate of around 70% at the desired false positive rate. I also developed a second version using a different machine learning model that achieved even better results at around 90%, but this second version was not fully productionalized. I built a Flask service and containerized the project using Docker.

Programmed a machine learning model to detect the presence of high-quality ID documents in an early stage of the pipeline. Despite initially having a dataset with more than 10% incorrect labeling, I achieved a true positive rate of 98% and a false positive rate below 0.001%.

To address the mislabeled data, I worked with the annotation team to reannotate the dataset and implemented strategies to overcome the incorrect labels. This project required strong problem-solving skills and the ability to work closely with a team to ensure accurate results.

Developed integration tests and tools for Ascom's message server to ensure 99.999% reliability. This included creating a script to simulate a large number of devices sending and receiving high loads using dedicated hardware and a simple Python simulator using multiprocessing. I also remotely controlled several Android devices using the Android Debug Bridge to verify performance and utilized pytest and Python to develop these tests and tools.

Analyzed simulated historical Starbucks data to determine what influences customers to make purchases and the best offers to send them as part of my data science nanodegree at Udacity.

Once I explored customer profiles, a transcript of events, and information on available offers, I found that offers were sent out in bursts, and transactions increased after an offer was sent. I also looked at customer spending habits and found that most customers spent around $20 per month, but because of outliers, the mean was much higher at $107, and the median was $72.

I analyzed the available offers and plotted their distribution among customers. Then, I used machine learning to build a model to predict whether a customer would make a purchase. The best-performing model was a gradient-boosting classifier with an 0.86 F1 score. I used this model to make recommendations on which offers to send to different segments of customers.

For more details, visit the project URL.

Applied my skills in deep learning and computer vision to build a model that classifies each pixel in an image into one of several categories. This project was a challenging and rewarding experience that was part of Udacity's Self-driving Car Engineer nanodegree.

While working on this project, I used the Python programming language and libraries, such as TensorFlow, to train a CNN to perform semantic segmentation. I worked with real-world data and optimized my model to achieve the best possible performance. This required experimenting with different network architectures, hyperparameters, and techniques, such as data augmentation and regularization.

Throughout the project, I focused on my learning and development, researching and studying various concepts and techniques related to semantic segmentation and applying what I learned to the project. By the end of the project, I developed a deep understanding of this topic and gained the skills and knowledge needed to tackle other challenging problems in the field of self-driving car technology.