Selected Work
World Cup Pass Detection
Computer Vision / Sports Analytics
Real-time system that detects players and the ball and predicts passes from World Cup footage, built on the Roboflow Sports repo. The launch demo reached 600k+ impressions on X, out-performing a concurrent Google DeepMind release on the same problem.
HackITBA 2026 · Plague Prediction
AI / AgriTech
1st place at HackITBA 2026, Buenos Aires' largest hackathon (IEEE Computer Society). An AI system that predicts corn leafhopper plague risk from historical, climate and satellite data, helping prevent crop losses that topped USD 3.6B in Argentina in 2023-2024.
NBA 3-Second Violation Detection
Computer Vision / Sports Analytics
AI system combining object detection, tracking, keypoint detection, pose estimation, zone occupancy detection, and time in zone counting per player. Featured in viral LinkedIn post with +2000 reactions.
Convolutional Kolmogorov Arnold Networks
Research / Deep Learning
Novel neural network architecture combining the Kolmogorov-Arnold representation theorem with convolutional neural networks. The paper has received more than 150 citations and was featured in IEEE Spectrum.
Satellites On Fire
Startup / Wildfire Detection
Real-time wildfire detection system using satellite imagery and deep learning. Founded at age 16, monitored +50 million hectares worldwide, secured $200k+ investment.
Super Mario DIAMOND
Neural Game Engine
Diffusion model trained to replace the Super Mario Bros game engine with learned dynamics. Generates game frames in real-time based on player actions.
Reading an LLM's Inner Thoughts
AI Safety / Interpretability
Built in 36 hours at Hack Buenos Aires (Platanus Ventures). An open-source system that uses Anthropic's Natural Language Autoencoders to read a model's inner thoughts, detect deception, and steer models back into alignment.
Open Source
Combined stars across repositories I author or contribute to
Experience
Junior Open Source Engineer
RoboflowAdvancing object tracking capabilities while pursuing thesis research on LLM memory optimization.
- -Implementing new trackers from state-of-the-art research papers
- -Developing and integrating different Kalman Filter variants for improved tracking accuracy
- -Implementing IoU variants for enhanced object association
- -Thesis research: LLM memory optimization via KV cache eviction strategies
- -Recent hackathon project: Using Natural Language Autoencoders (Anthropic research) for LLM alignment
Open Source Intern
RoboflowContributed to open-source computer vision libraries and created educational content for the AI community.
- -Developed an AI system to detect 3-second rule violations on NBA videos, combining object detection, tracking, keypoint detection, pose estimation, and zone occupancy detection (+2000 reactions on LinkedIn)
- -Contributed to the Trackers library (2k+ stars), implementing object tracking papers from scratch
- -Contributed to Supervision repository with over 33k stars, the main library on computer vision tools for developers
- -Created dataset with automatic labelling pipeline and hand revision, trained CV models to classify player air/ground status
- -Added Huggingface's SmolVLM to Maestro library for finetuning multimodal models (+2.5k stars)
- -Developed the basketball section on Roboflow sports repository (+4.5k stars)
- -Added model evaluations to Roboflow's model leaderboard for SOTA object detectors
- -Published technical blog posts explaining computer vision algorithms
- -Finetuned VLMs for reading basketball jersey numbers
Founder & CIO
Satellites On FireFounded AI startup for real-time wildfire detection at age 16. Built systems that helped protect communities and natural resources.
- -AI algorithm monitored +50 million hectares all around the world
- -Secured $200k+ in investment funding
- -Company raised $2.7M USD three years after my departure
- -Developed proprietary image processing algorithms for satellite data
- -Awarded Scientific and Cultural Interest by Buenos Aires Congress
Writing
SORT Explained: Real-Time Object Tracking in Python
Comprehensive guide to the SORT algorithm for real-time object tracking in computer vision applications.
Detect 3-Second Violations with AI in Basketball
Building an AI system to detect NBA rule violations using detection, tracking, and pose estimation.
Gradient-Based Methods in Machine Learning
Deep dive into the mathematics behind gradient descent and optimization algorithms in ML.
Serverless ML with GPUs on Google Cloud
Deploying ML models with GPU acceleration using Cloud Run for cost-effective inference at scale.
Contact
Interested in collaborating on computer vision, open source, or research projects? Let's talk.