Ehran Lenaerts

PAST PROJECTS

SELF DRIVING CAR SIMULATION

During my AI course, my final group project focused on developing a simulated self-driving car using Udacity and Python. This project allowed us to explore two distinct approaches: one utilizing OpenCV and the other employing a neural network. In this challenging endeavor, we aimed to replicate real-world autonomous driving scenarios within a simulated environment. Through our diligent efforts and collaborative teamwork, we crafted a comprehensive solution that showcased the potential of AI in the field of autonomous vehicles.

APPROACH

PHASE 1

Firstly, we embarked on the OpenCV approach, leveraging the power of computer vision techniques to analyze and interpret the simulated car's surroundings. By implementing image processing algorithms, object detection, and feature extraction, we aimed to simulate the car's perception capabilities and enable it to make informed decisions based on the visual information.

PHASE 2

In the second approach, we delved into the realm of neural networks. Utilizing deep learning techniques, we developed a neural network model capable of learning from the car's simulated driving experiences. By training the network on a vast dataset of labeled images, we sought to equip the self-driving car with the ability to make predictions and navigate through the simulated environment autonomously.

OBSTACLES

There remained a few challenges and obstacles in our path however. Obtaining high-quality training data proved crucial, and we employed innovative techniques to preprocess the data accurately. Another hurdle was accurately detecting and calculating road boundaries, which we addressed through the use of Hough Lines and fine-tuning parameters. Training the model posed difficulties due to shadow interference, requiring us to employ data augmentation and preprocessing methods to enhance the model's robustness. Despite these obstacles, our dedication, adaptability, and problem-solving skills enabled us to overcome these challenges and deliver a successful self-driving car simulation.

PERSONAL ROLE

In the project, my primary focus revolved around the OpenCV aspect, with a key emphasis on optimizing our video pre-processing techniques. I strived to ensure that our non-AI version of the self-driving car functioned effectively, exploring methods such as Hough transformation and predictive algorithms based on previously detected data. Additionally, I played a crucial role in linking our control scheme to the output of OpenCV, facilitating smooth integration and seamless functionality of the overall system. My contributions encompassed both technical expertise in computer vision and a deep understanding of the interplay between data processing and control mechanisms, ensuring the success and functionality of our self-driving car simulation.

RESULTS AND ACHIEVEMENTS

In conclusion, our self-driving car simulation project was a journey filled with challenges and valuable learning experiences. Throughout the process, we successfully tackled numerous issues and expanded our knowledge in various domains, including neural networks, OpenCV image processing, flask integration, and data cleaning. By working through the intricacies of neural networks, we gained a deeper understanding of their capabilities and limitations, allowing us to optimize our model's performance. The application of OpenCV image processing techniques enabled us to extract valuable insights from visual data, enhancing the car's perception and decision-making abilities. Beyond technical skills, the experience of working as a team in this self-driving car simulation project honed our interpersonal skills, such as conflict resolution, adaptability, and time management. Overall, this project served as a stepping stone for acquiring practical knowledge and skills in key areas such as neural networks, image processing, web integration, and data management.

RESULTS AND ACHIEVEMENTS

Project Paper

HEADLESS E-COMMERCE WEBSITE

One of the projects I had the opportunity to work on was the development of a headless e-commerce site for Elysion, an innovative company in the industry. Collaborating in a team setting, we undertook the task of creating a seamless and engaging online shopping experience for their customers. With a focus on flexibility, scalability, and cutting-edge technologies, our goal was to deliver a robust and customizable e-commerce solution that aligned with Elysion's brand and business objectives.

APPROACH AND METHODOLOGY

Throughout the project, we embraced the concept of a headless architecture, which decouples the frontend and backend systems, allowing for greater flexibility in design and development. This approach enabled us to create a dynamic and interactive user interface while leveraging the power of APIs and backend services to handle business logic, inventory management, and order processing.

My team collaborated closely with Elysion to understand their unique requirements and translate them into tangible features and functionalities. We worked diligently to design and implement a visually appealing and user-friendly interface that provided customers with a seamless browsing, selection, and checkout experience.

PERSONAL ROLE

Within the project of developing a headless e-commerce website, my role encompassed various responsibilities in a DevOps capacity. I played a crucial role in ensuring the seamless integration and collaboration of different components, encompassing the front-end, back-end, and hosting capabilities. This involved implementing robust deployment pipelines, configuring continuous integration and delivery systems, and orchestrating the smooth interaction between different services. Additionally, I helped with setting up the Algolia integration, which enabled our website to effectively track visitor behavior and gather valuable data for analysis. By implementing this feature, we could optimize user experience, improve search functionality, and provide personalized recommendations to enhance customer engagement. Furthermore, a significant focus of my contributions revolved around fully integrating Adyen, a versatile payment service provider. I worked diligently to ensure a flawless checkout experience for customers by implementing secure payment methods, seamless transaction processing, and comprehensive error handling. This involved configuring and customizing the Adyen integration, adhering to industry standards, and conducting rigorous testing to ensure smooth payment transactions.

RESULTS AND ACHIEVEMENTS

Working in a collaborative team environment, we leveraged our diverse skills and expertise to successfully deliver the headless e-commerce site for Elysion. Through effective communication, agile project management, and a shared commitment to excellence, we achieved our goal of creating a powerful, customizable, and customer-centric online shopping platform.

Sadly enough, the site was taken offline to preserve resources within the company.

Final presentation

Final documentation

BIG CAT CLASSIFIER

During my academic journey, I had the privilege of collaborating on a captivating group project as part of a school task. Our collective goal revolved around the development of an advanced image classifier, meticulously designed to discern and classify various species of big cats. Delving into the world of wildlife, we specifically selected big cats as our subject matter due to the fascinating nuances and unique characteristics that distinguish each species. The choice of big cats presented us with an engaging and complex challenge. Their diverse physical attributes, such as size, coloration, and patterns, demanded an image classifier capable of accurately differentiating between them.

APPROACH

PHASE 1

To create the image classifier, we employed transfer learning techniques using the ResNet34 architecture. This pre-trained model, which is trained on the large-scale ImageNet dataset, provided a solid foundation for our project. By fine-tuning the model with our dataset of big cat images, we were able to train a convolutional neural network (CNN) model that achieved impressive classification performance.

PHASE 2

To make the image classifier accessible and user-friendly, we deployed it using Streamlit, a powerful framework for building interactive data applications. With Streamlit, we created a simple and intuitive interface that allowed users to upload their own images and obtain instant predictions on the species of the big cat in the image. In addition to the classifier, we also implemented an exploratory dataset page within the Streamlit application. This feature allowed users to explore the dataset we used to train the model, providing transparency and insights into the training process. Furthermore, we incorporated a training functionality within the Streamlit GUI, enabling users to train a new model using their own custom dataset. This feature empowered users to experiment with different training approaches and further explore the capabilities of deep learning models.

COMPARISON

As part of our project's learning objectives, we also tested the trained model on Google's Teachable Machine, which is a web-based platform for training and deploying machine learning models. The results obtained from Teachable Machine were comparable to our own classifier, demonstrating the performance of the model.

PERSONAL ROLE

Within the context of the Big Data project, my personal role encompassed several crucial responsibilities. Firstly, I took charge of acquiring the dataset by employing a web scraping technique to gather the required data. This involved leveraging my technical skills to extract relevant information from various online sources and compile it into a structured dataset. Subsequently, I undertook the intricate task of data cleaning. Recognizing the importance of high-quality and reliable data for accurate analysis and modeling, I dedicated significant time and effort to ensure that the dataset was free from inconsistencies, errors, and duplicates. I also played a significant role in setting up the training configuration for the image classification model itself. Recognizing the importance of a well-configured training process for achieving optimal model performance, I helped to define and fine-tune the various parameters and hyperparameters involved in the training phase. In addition to data acquisition, model training and cleaning, I worked on configuring the Google Teachable Machine. Furthermore, I actively participated in conducting rigorous testing and evaluation between our developed classification model and the Google Teachable Machine. This comparative analysis allowed us to gauge the performance, accuracy, and effectiveness of our model in relation to an established industry benchmark. Through this process, we gained valuable insights into the strengths and weaknesses of our approach and identified areas for improvement.

RESULTS AND ACHIEVEMENTS

During our project, we explored various techniques to classify images using pre-trained models. We were thrilled with the excellent results achieved by our team, but it's possible that the chosen subject of classification, big cats, made the task somewhat easier. To further enhance our project, it would be fascinating to delve deeper into the performance of differently trained models available in the fastai library. This could provide insights into the comparative effectiveness of various training approaches and help us expand our understanding of image classification. Additionally, we could consider adding more cat types that bear resemblance to each other, such as the jaguar, to challenge our model further. This would allow us to evaluate its robustness in distinguishing between closely related species and assess its adaptability to more complex classification scenarios. Overall, our investigation into image classification using pre-trained models was a success, yielding impressive results. However, there are exciting opportunities for future research and experimentation to expand the scope of our project and deepen our understanding of machine learning techniques.

RESULTS AND ACHIEVEMENTS

Try it out!

ME-ARTIST GALLERY

For a personal side project, I took on the task of assisting someone in hosting their own website to showcase their artwork in a comprehensive manner, while also incorporating the music that is intricately linked to their pieces. The individual had a specific preference for a minimalist design, which presented an opportunity for me to emphasize functionality and prioritize seamless hosting. The project involved crafting an intuitive user experience, ensuring that the website not only showcased the artwork effectively but also provided a seamless integration with the accompanying music. By focusing on essential functionalities and optimizing the hosting process, the objective was to create a clean and streamlined platform that allowed the artist's work to shine.

APPROACH

GENERAL LOOK

In this project, the artworks are organized into distinct categories, creating a structured presentation of the pieces. The website layout is designed to showcase the artworks in a visually appealing manner, with each piece elegantly displayed underneath one another. The dynamic nature of the layout ensures that the artworks adapt to the viewer's screen size, providing an immersive and enjoyable viewing experience. To enhance the connection between the artworks and the artist's musical inspirations, each piece is accompanied by its title and the corresponding song title that should be played while viewing it.

ADMIN BACK-END

In addition to the website's aesthetic presentation, I implemented a user-friendly admin tab specifically designed for the artist. This intuitive interface allows the artist to easily add new categories and art pieces to the website without the need for external assistance. The added functionality is seamlessly connected to a database, ensuring that any changes or additions made by the artist are automatically reflected and updated on the site. This streamlines the process of managing and updating the website's content, providing the artist with full control over their portfolio and enabling them to showcase their latest creations effortlessly.

RESULTS AND ACHIEVEMENTS

In conclusion, although the project was initially undertaken as a practice endeavor to assist someone in showcasing their artwork, the final result turned out quite nice. The website's visually appealing design, combined with its user-friendly platform, provides a seamless and captivating experience for visitors to explore and appreciate a variety of artworks. The successful implementation of features such as dynamic display and an intuitive admin tab for content management highlights the project's achievement in delivering an easy-to-use and aesthetically pleasing platform for showcasing artwork. Overall, the project can be deemed a resounding success, fulfilling its purpose of providing a comprehensive and visually captivating platform for sharing and appreciating art.

RESULTS AND ACHIEVEMENTS

Take a look yourself!