My journey began with a passion for Mechatronics, which led me to pursue a degree at The Hague Applied Sciences with a specific focus on machine learning, robotics, and computer vision. During my academic journey, I was fortunate enough to work part-time at a wireless mesh sensor company, where I gained invaluable practical experience.
After my graduation, I embarked on a pre-master’s degree at TU Delft to further my knowledge in computer engineering. I’m proud to say that I completed my pre-master’s while working part-time, and subsequently finished my Computer Engineering Master’s program in just 1.5 years instead of the standard 2-year timeframe.
My thirst for knowledge and desire to make a meaningful impact led me to join Experience Data, a leading Data Science company, and its subsidiary, EFQ. As a key member of the team, I take great pride in using my skills and expertise to deliver top-notch solutions to our clients. (And maybe even pick up a few grey hairs in the process!)
As a data science consultant at my current company, I have the privilege of working on a diverse range of projects. But my current focus is on making machines smarter and more efficient with the power of AI.
One of my recent projects involved designing and building a fruit sorting machine that leverages the latest AI techniques to enhance the quality control process. Using my expertise in machine learning, I developed an advanced image model that enables the machine to classify and sort fruits with unparalleled accuracy.
I strongly believe that making AI smarter and smarter is dependent on the quality of labeling the data. As a result, I have invested significant time and effort in developing robust labeling techniques that enable machines to learn and adapt with greater precision. By improving the quality of labeled data, we can unlock new levels of accuracy and performance, ultimately leading to more effective and efficient machines.
For me, the most fascinating aspect of this work lies in the intersection of engineering and data science. By integrating physical objects with cutting-edge technology, we can create machines that are not only smarter, but also more responsive to their environment. And with the power of AI, we can unlock new levels of efficiency and productivity that were previously unimaginable.
Software skills
The table below provides an overview of the programming languages that I am proficient in. My primary focus is writing code for microcontrollers in C/C++ and for desktop/embedded computers in Python. However, my experience extends to web development using a variety of technologies such as NodeJS, Angular, Django, and FlaskAPI. Most recently, I have been using FastAPI to develop web applications. I have also worked with React, building small pagination tables and buttons that trigger remote devices. Additionally, I have experience with PostgresSQL and have had to create complex queries that needed to be executed quickly. In the realm of data science, I primarily use Pandas and NumPy in Python and have worked with various AI libraries such as Scikit-learn. To aid in visualization and research purposes, I have utilized Power BI. My knowledge of diverse programming languages and frameworks allows me to approach projects from various angles and find the most effective solutions.
| Advance | Experience | Beginner |
|---|---|---|
| C | Makefiles | Rust |
| C++ | Linkerscripts | PHP |
| Python | Matlab | |
| Bash | ||
| Angular (>V2) | ||
| Node.js | ||
Databases:
|
||
| VHDL | ||
| Verilog | ||
| Writing SRS |
In one of my projects, I developed a custom framework that uses a streamlined makefile with consolidated linkerscript files to facilitate fast compilation of the code in mere milliseconds. This approach is in contrast to the use of large libraries that are slow and can consume a lot of space in the limited flash memory available on these devices. In one project, I even had to rewrite standard input/output library (stdio.h) to make it fit within the memory constraints of the microcontroller.
Printed Circuit Boards (PCBs)
I started circuit board making when I wanted a board with fpga and 2 microcontrollers for a home project. After that, I developed different PCBs at a company I worked at. One of the projects was making a PCB tester with pogo needles so that you can quickly test if the complex PCB passed the tests. After that, I created different sensing PCBs like a water level meter with LoRa communication. I have done some FPGA and MCU PCBs, where the other MCUs were programmed for MCUs, an ESP32 WiFi sensing board like the one that is controlling my own central heating system, and even a light panel that had special wavelength LEDs for measurements purposes. I developed more than 20 different PCB designs (and counting). One of the latest projects is a microcontroller/FPGA debugger, which I wrote about in one of my blog posts.
Artificial Intelligence
My career has mainly revolved around the exciting and constantly evolving field of Artificial Intelligence, with a focus on Machine Learning, Deep Learning, and Spiking Neural Networks. Whenever I start working with a new dataset, my first step is to validate its performance and ensure that it is presented in the most effective way possible. I also take great care in determining the model’s requirements and what levels of robustness and performance are necessary, while always questioning whether AI is truly the best solution. With so many methods and techniques to choose from in the vast field of AI, I have honed my skills in a number of areas and created a small list of methods that I have worked with.
| Machine learning |
|---|
| MLP |
| SVM |
| Cross-validation |
| n-fold rule |
| Descision trees |
| Adaboost, Bagging, Stumping |
| Model reduction |
| Regressing and PCA |
| Clustering methods like Meansift, SOM, K-means, and others |
| Viola and Jones |
| Learning methods: Momentum, Rprop, Adams |
| Activation functions |
| Novelty detection |
| Heat-maps and other data validation techniques |
| Small evolutionary algorithm with fitness function |
| Time-series datasets |
| Normalization techniques |
| Deep learning |
The following list shows a more indepth view of deep learning.
| Deep learning |
|---|
| RESNet and variants |
| VGG and variants |
| Inception and variants |
| Attention learning |
| GANs |
| LSTMs |
| BabyRNN |
| C++ implementation of a CNN |
| Framework Keras (see paper-blog for source code) |
The following list is for spiking neural networks.
| Spiking Neural Networks |
|---|
| digital ANN to analog SNN |
| Rate encoding and spike alorithms |
| Tempotron, SpikeProp, and SpikeTemp |
| Leaky-integrate and fire neuron model |
| Liquid State Machine |
Education master topics
Throughout my master’s program, I had the opportunity to explore a diverse range of topics:
- Network Security (Parts of Advance Network Security and crypto)
- Deep Learning, Computer Vision
- CPU design: caches, core, ALU (like a 2-cycle multiplier)
- GPU, FPGA, Multi-data instructions
- Operating systems
- ADHOC networks focus on mesh-networks
- Web data management
- Entrepreneurship, business plan