Hi, I’m Oriol Gorriz, nice to meet you 😄.

Please check out my LinkedIn for more information.

👨‍💻 I'm a spanish guy who started programming very young and has a lot of passion for the potential that software has to revolutionize society.
🏠 I live in a town near Barcelona (Spain).
🚴‍♂️ My favourite sport is cycling.
🚗 Experience with computer vision applied to cars and path planning. 🧠 I'm very interested in autoregressive Language Models and the future architecture changes that might enable AGI.

After completing a Baccalaureate oriented in science and technology with a 8.67/10 final grade, my final work already aimed to a specific field of study. The work was about controlling a robotic arm using a Microsoft Kinect programmed using Processing and Arduino (grade 10/10).

Then, I started a bachelor's degree in Computer Engineering at UPC (Technical University of Catalonia, 240 ECTS).
I could not have chosen better, I loved everything we did!
There, I learned:

• To program appropriately (C++)
• How a CPU works
• Communications between computers
• Assembler programming
• Designing and structuring complex software programs
• Algorithmics
• Data structures
• And a lot more!

I decided to study a master's degree in Data Engineering and Analytics at TUM (Technical University of Munich, 120 ECTS). My decision came from the realization that AI is essential to create full, true intelligent systems. The key learnings are:

• Classic & State-of-the-art Natural Language Processing
• Reinforcement Learning
• Probability Theory
• DL Methods for Robotics
• Hands-on projects in robotics and LLMs

My semester thesis is about neural image compression. By conditioning compression not only on the input image but also on its object features, our method achieves a lower distortion on Regions of Interest (ROIs), effectively shifting bitrate from non-ROIs to ROIs. This is particularly relevant in teleoperated driving imagery. See the repo.

My master's thesis optimizes the load balancing of distributed LLM inference. Our framework is built on top of llama.cpp and boosts generation throughput on heterogeneous industrial edge devices.

While studying, I decided to join my university's Formula Student team (BCN eMotorsport). There I learnt to plan, develop and finish projects and the importance of a well-coordinated team. Later on, I became the head of the perception department of our driverless car. My job was to ensure that our perception system met the requirements in order to run as fast as possible while developing components for the perception software stack.

My bachelor's thesis consisted of two systems that later were used in the car during the competitions:

• One of the proposed systems CCAT is about creating a statistical model to sensor fuse the cameras and the LiDAR in order to obtain classified cones and keep track of them.
• The other proposed system Urimits is a track limits detection algorithm that only takes geometrical data (position) of every cone in track to compute the track limits.

After completing my thesis, I continued working for the team and I programmed Urinay, a much smarter track limits detection algorithm that takes a completely different approach than Urimits and gives much more appealing results leveraging Delaunay triangulation and an efficient implementation of beam search on the triangles' edges.

During my master's degree, I joined TUfast Racing Team, the Formula Student team of TUM. There, I developed a DBSCAN-based cone detector from a multi-scan global map, which we used during competitions (alongside Urinay) and allowed us to win the driverless competition in Spain.

From the experience obtained at university and in the Formula Student team, my main interest are driverless vehicles. Autonomous driving has a lot of advantages compared to manual driving and is my desire to contribute to society with advancements in this field. I am willing to work in the perception of vehicles, how can they see the world, detect objects and know what to do in EVERY situation? This is the challenge.

While I was in university, in my free time, I decided to learn how to program an application properly. That is because while I was in high school, I developed two applications using App Inventor and it leaved a bitterwseet taste because it did not provide all the functionality I was looking for.

So I learned Flutter.

With some of my friends from university we decided to make an app for students to make groups in class to work. It often happens that new students work alone since they do not know anyone to work with. This app FibGroup was developed during a hackathon called HackUPC (the largest in Spain). The app allowed students to navigate through different subjects and classes, then they can either create a group or join one (each group has a pre-defined number of people). Then the people already in the group can accept or decline new requests.

Some time after, I decided to make an app (more complex) for my Formula Student team to help with team management. All info can be found in BCNeMotorsportApp.