We can organize an interview with Aldin or any of our 25,000 available candidates within 48 hours. How would you like to proceed?
Schedule Interview NowMy journey at Softaims has been defined by curiosity, growth, and collaboration. I’ve always believed that good software is not just built—it’s carefully shaped through understanding, exploration, and iteration. Every project I’ve worked on has taught me something new about how to balance simplicity with depth, and efficiency with creativity. At its core, my work revolves around helping businesses and people achieve more through thoughtful technology. I’ve learned that the most successful projects come from teams that communicate openly and stay adaptable. At Softaims, I’ve had the opportunity to work alongside professionals who challenge assumptions, share knowledge generously, and inspire continuous improvement. I take pride in focusing on the fundamentals—clarity in logic, consistency in design, and empathy in execution. Software is more than a set of features; it’s a reflection of how we think about problems and how we choose to solve them. By maintaining this perspective, I aim to build solutions that are not only effective today but also flexible enough to support the challenges of tomorrow. The culture at Softaims promotes learning as an ongoing process. Every new project feels like a step forward, both personally and professionally. I see each challenge as a chance to refine my skills and contribute to the shared vision of building technology that genuinely improves lives.
Main technologies
2 years
1 Year
1 Year
1 Year
Potentially possible
Prediction of GOOG Stock Market with LSTM and Web Application with Flask. LSTM isn't a good algorithm to predict a lot of days ahead of a stock market, but the intention of this project was doing an article on medium where I could explain how to simple build web applications.
The project was based on an X-ray teeth images dataset, where I have the input images (X-ray) and the segmentation masks (outputs). Each tooth is represented with a different color in grayscale. All the project was made with TensorFlow Keras. The U-Net model is mainly used for image segmentation tasks because of it's good performance on small and large datasets. If you have a very small dataset, you can also fine-tune your model by retraining a model that was originally trained on a larger dataset for a similar task. This is beneficial because you can freeze some layers during the retraining process, allowing you to avoid overfitting and reduce computational costs without compromising accuracy. For the training, I applied augmentation by rotating and slightly zooming the images to enhance accuracy on the test dataset predictions. I used a batch size of 4, ran 10 epochs, set the learning rate to 0.001, and specified a height/width of the output image as 512 pixels with 33 filters (considering that a person has 32 teeth, and accounting for the 0 pixel values, the total is 33 filters).
This is just a fun project that I made to predict whether bodybuilders are doing the correct poses in the category (just double biceps until now). This project reached more than 3000 likes on Linkedin. It can extend to a world where this technology is on all gym mirrors and "normal" people could pose in front of them, and the result shows at the same time. OpenCV and MediaPipe give us the power to explore the world of computer vision, but we also can do it with segmentation, for instance. It's that thing: "GIVE ME 10,000 LABELED IMAGES, AND WE TALK MORE LATER."
Other in Computer science
2018-01-01-2022-01-01