Portfolio
Software engineer focused on applied AI, product-oriented ML systems, and computer vision tools. Currently finishing AI Engineering at UdeSA.
Based between San Francisco and Buenos Aires. Open to AI/ML research and product engineering roles.
Selected work
Full-stack product work, applied ML/AI, robotics, RL, and a few side projects.
Chrome extension that filters bots, spam, and AI slop on X using custom ML pipelines. Built across a TypeScript extension frontend, FastAPI services, Redis queues, Dockerized backend, Supabase storage, and custom detection models.
Reinforcement learning final project: a plain PyTorch PPO-Clip implementation trained from scratch on CarRacing-v3, plus experiments on how aggressively the visual state can be compressed while still learning effectively.
Wall-mounted pen plotter built from scratch with stepper motors, belt drive, and custom control electronics. Iterated from Arduino Nano to ESP32 plus Raspberry Pi and used it to produce large-format plotted drawings.
Interactive data story about unregulated Chinese fishing near the Argentine EEZ. Won second place in a 2025 national data visualization competition organized by Argentina's science and innovation secretariat.
Webcam-based eye gaze prediction system trained on a custom multi-user dataset with varied head poses and lighting. Benchmarked ridge baselines against CNNs, ResNets, and MobileViT models and beat WebGazer-style regression.
Thesis subproject focused on pedestrian detection from drone viewpoints. Fine-tuned RF-DETR on VisDrone to outperform the quality of readily available open-weight YOLO baselines for aerial pedestrian detection.
Weekend hackathon research project where an RL agent learns to doomscroll 'optimally' by maximizing a synthetic brain-activation reward predicted by Meta's TRIBE v2 fMRI Transformer. Built across reward modeling, PPO training runs, replay/export tooling, and a public Next.js demo.
Emergency drone landing system for dynamic urban scenes. It predicts obstacle motion, builds a spatiotemporal risk map, and selects the safest landing point and touchdown time under UAV motion constraints.
Implemented FastSLAM for a Turtlebot, combining probabilistic localization and mapping into an end-to-end robotics project.
Trained a Deep Q-Network from scratch in PyTorch for the Centipede environment, focusing on the fundamentals of value learning rather than library abstractions.
About
I'm a software engineer currently finishing a degree in AI Engineering at Universidad de San Andrés, in Buenos Aires. My work focuses on applied AI systems that are both technically serious and product-oriented. I love robotics, reinforcement learning, and agentic systems. I also have a good eye for visual storytelling and love photography.
Education
Buenos Aires, Argentina. Expected graduation: July 2026. Part of the program's first graduating cohort.
Completed three years of a five-year degree before switching into AI Engineering.
Career
Worked on backend data infrastructure and product features at a Google Trends intelligence startup whose Chrome extension served roughly 100k users. Built internal tooling, user-facing features, and led a $100k/year hedge fund client project. Also handled ops.
Early hire part-time contributor at a seed-stage startup focused on people search and serendipity. Supported product and data workflows including scraping, parsing, data cleanup, analysis, and other early-stage operational work.
Contacts
Based in SF / Buenos Aires and currently focused on applied ML/AI roles. Email is the fastest way to reach me.
