🗣️ Talks

Live-built a fully functional generative AI product in 30 minutes, zero lines of code — a custom avatar bot using Telegram + Make.com + HuggingFace Spaces. Part of the 9th PyData Granada Meetup, held within the II Congreso de Inteligencia Artificial de Andalucía. The talk challenges the assumption that shipping AI products requires deep technical knowledge: with the right low-code tools, anyone can prototype fast.

A live experiment at TechFest 2024, one of Spain's most prominent tech festivals. I tracked a real basketball in 3D using locally-running computer vision models and visualised everything in real-time with Rerun — no cloud, no pre-recorded video, all on-device. The audience saw the 3D reconstruction appear on screen as the ball flew through the air. One of the talks I'm most proud of.

The first public run of the 3D computer vision basketball tracking demo, presented at the Spain AI Málaga chapter meetup. Same concept that later became the TechFest 2024 talk — live multi-camera 3D reconstruction of a moving ball, shown on screen in real-time.

Video not recorded for this session.

Developing inside a Docker container on VSCode gives you full control of your environment — not just Python dependencies but also the OS and system requirements. This makes it trivially easy for colleagues to start developing in the same repo. Since I discovered this tool I've been using it for every single project. I gave this talk to my colleagues at Sngular 🎉 and they loved it 💙.

No recording available for this session.

The culmination of a side project: detect facial expressions and show the complete pipeline — data acquisition through face key-point detection, normalization to suppress rotation, data labeling (normal, angry, happy, sad, surprised, winking), training a k-NN classifier and live webcam inference. This talk was very interactive since anyone could use the project live.

As organizer of the Malaga AI association, I conducted an introductory course on deep learning applied to robotics (Python + TensorFlow + Docker) for 8 lessons over 2 months. Students from diverse engineering fields learned to set up their IDE, Python basics, trained an RL agent for CartPole using TensorFlow, and saw how this applies to robotics with ROS. The most challenging course I've given — and the one I learned the most from! 🤓

I took hundreds of pictures of Mocka, one of my cats 😂, to show at TabularConf 2021 (my first national congress 💪). Using the ImageAI library, I trained a YOLOv3 model in a Jupyter Notebook and showed how object detection models are validated using the mAP metric.

With my friend Quino and the Spain AI Association, we ran a 6-lesson introductory course on Python basics and Reinforcement Learning. Thanks to Covid, we moved it online, reaching up to 200 students per class. 🎉 Students trained an RL agent using OpenAI Gym + TensorFlow to solve the CartPole problem.

Covid hit and the world stopped — but impressive movements emerged. The CoronaConf brought programming knowledge to everyone. I volunteered to give an intro to Generative Adversarial Networks, drawing from the magnificent book Generative Deep Learning by David Foster.

While working on my Master's thesis on GANs I gave a gentle introduction to Generative Adversarial Networks at Demium. In one hour I covered the main concepts of GANs, live code with TensorFlow, and practical near-future applications. Still remember the Will Smith "I, Robot" reference 😄

My first ever talk! I spoke about CNN architectures — how convolutional layers work, how to train them, selecting hyperparameters, and showed TensorFlow code. I did it in English (what an achievement 💪) for the School of AI association that me and friends started months before. I was terrified, but it went great!