Hi, I'm AlexanderBodner

Satellites On Fire was my startup that provided real-time wildfire alerts using satellite and camera imagery. As the founder and CIO, I led the development of AI models capable of detecting wildfires from various image sources. The system used a serverless AWS backend to process images, run detection algorithms, and send instant notifications to users. The startup secured over $400,000 in investment and is still active, but I made a successful exit. This project combined my passion for AI with a solution to a critical environmental problem.

In this research project, I proposed a novel architecture for Convolutional Neural Networks based on Kolmogorov-Arnold Networks applied to Computer Vision. The key innovation was designing kernels as functions of the input pixels, enabling more learning at each convolution step. I published a paper on this work and released the open-source code, which has garnered more than 800 stars on GitHub and over 75 citations from other researchers. This project demonstrates my ability to contribute to the theoretical advancement of deep learning while providing practical implementations for the research community.

During my internship at the Robotics and AI Lab at UdeSA, I built a self-balancing robot with two wheels and developed a mobile application to control it remotely. The robot featured a camera system that enabled it to detect, track, and follow objects of interest. This project combined principles of control systems, computer vision, and mobile app development. The robot demonstrated stable self-balancing capabilities while simultaneously processing visual information to track and follow targets, showcasing the integration of multiple technologies in a single robotic system.

In this project titled '¿Sueñan los androides con Marios eléctricos?' (Do androids dream of electric Marios?), we adapted the DIAMOND architecture to create a neural network that replaces the Super Mario Bros game engine. The system consists of two main components: a Denoiser that predicts future game states in low resolution (64x60 pixels) and an Upsampler that increases the resolution to 256x240. We experimented with various data collection methods, including reinforcement learning agents and human gameplay, to train the model. The best results came from human gameplay data and balanced artificial datasets with game insights, which allowed the model to respond accurately to player inputs while maintaining visual fidelity and temporal coherence. This work demonstrates how diffusion models can be used to simulate complex game dynamics without traditional game engines.

How Much VRAM is a web-based calculator that helps AI researchers and engineers estimate the GPU memory requirements for training and inferring with deep learning models. This tool addresses a common pain point in the machine learning workflow: determining if a model will fit on your GPU before starting a potentially time-consuming training process. The calculator supports both PyTorch and TensorFlow frameworks and allows users to input various parameters such as model architecture, batch size, input dimensions, and precision (FP32, FP16, or INT8). By providing accurate VRAM estimates, the tool helps users optimize their model configurations and avoid out-of-memory errors. The project has gained popularity in the AI community, achieving over 100 stars on GitHub, and serves as a valuable resource for both beginners and experienced practitioners in the field.

Trai is a Software-as-a-Service (SaaS) platform that enables online shoppers to virtually try on clothing items before making a purchase. Using advanced deep learning models deployed on Google Cloud Functions with GPU acceleration, the platform can accurately render garments onto customer photos while maintaining lighting, pose, and fabric details. The serverless architecture ensures scalability and performance while keeping costs efficient. This project demonstrates my ability to build production-ready AI solutions that solve real business problems in the e-commerce space.