2024
Next Generation Award
Society for Neuroscience (SfN)
For the podcast A Todo Neuro for the Argentinian Neuroscience Society.
Brands and institutions I've worked with
Doctoral neuroscience now turning into LLM interpretability work. From single-cell calcium signals in zebrafish to spatial reasoning in 70B-parameter models.
Tests whether multimodal LLMs that classify images from a few in-context examples can also explain their own decisions. Forcing formally structured, concept-based explanations degrades accuracy monotonically (93.8%→90.1%): strong few-shot visual classifiers, but not yet tuned for machine-verifiable explanations.
Tracks LLM internal states across conversation by combining concept probes with model self-reports. Shows that logit-based self-reports causally couple to probe-defined states, with the effect scaling up to R²≈0.93 on Llama-3.1-8B.
Maps abstract spatial models inside LLMs during a grid-world navigation task. Cartesian text representations win on success rate, and intermediate-layer probes reveal robust spatial features that transfer to unrelated reasoning tasks.
Whole-brain calcium imaging in zebrafish links tectum-level audiovisual integration to premotor circuit activity and behavioral enhancement. Custom rig, custom tracking pipeline, single-cell-resolution analysis.
Shows that goldfish integrate visual looms with brief auditory stimuli to raise C-start probability and shorten escape latency. Multisensory enhancement is inversely correlated with stimulus salience; a soma-integration model accounts for the behavior.
Config-driven Python library for training and using concept probes inside open-source LLMs. Builds and validates per-layer probes for any concept you can label, runs steering experiments, and produces the heatmaps and isotonic-R² metrics behind my interpretability papers.
Calcium imaging of audiovisual danger-stimulus integration in the zebrafish tectum and Mauthner cell. New experimental setup, custom tracking software, full analysis pipeline.
Society for Neuroscience (SfN)
For the podcast A Todo Neuro for the Argentinian Neuroscience Society.
Ministerio de Ciencia · Argentina
For environmental science shorts at Xplora.
Travel Grant
3-month research stay in Paris with Germán Sumbre — IBENS, École Normale Supérieure.
Fulbright Commission
Scholarship for studies at the University of Alabama, USA.
City of Buenos Aires
City of Buenos Aires academic-excellence award.
Code I've shipped, open and used by other people.
Python framework for orchestrating multi-agent LLM pipelines — chatbots, decision systems, and tool-using agents. Powers the bot on this page, the CasanChef sous-chef on WhatsApp, and a number of internal automations across TOXI Media's clients.
Read the docs →
Config-driven library for training, validating and steering concept probes inside open-source LLMs.
Multimodal AI products and brand experiences built end-to-end — from prompt engineering and fine-tuning down to the deployment your users actually see.
Multimodal AI sous-chef. Multi-agent system that finds personalized recipes from a user's pantry, on a chat interface deployed to thousands of users.
First Beauty-Tech experience in Argentina. Fine-tuned Stable Diffusion + LoRAs for realistic, instant hair-transfer in a brand-engagement mirror.
A daily LLM-generated dashboard analyzing Argentine economic and political indicators. Structured output, runs every morning, public.
Node-based AI video production workspace. Prompts, references and generated shots become a real storyboard, with a public read-only demo.
A curated player for AI-generated music. Albums, artwork, and a Spotify-style listening experience built from end-to-end synthetic content.
Multi-tool agent on Telegram. Calendar, Gmail, Spotify, vision-based meal logging, audio transcription. Real LLM, mocked tools in the public demo.
28 language pairs, contextual breakdowns, conjugations, IPA. Vanilla JS frontend, FastAPI + Groq backend. Public demo with real LLM.
Daily news report. Scrapes a curated X feed with Playwright, summarizes with Groq, pushes to Telegram. Demo varies by day-of-week.
Animated music videos, scripts, and broadcast content — produced in-house or as creative-tech direction for partner studios.
Animated, AI-assisted music videos of the Argentine national anthem and other classics — for kids, in production with a small team.
Animated kids' music videos. Original songs, full audiovisual production, AI-assisted animation for a recurring kids audience.
An animated children's series celebrating the traditions of Entre Ríos, produced end-to-end with El Perro en la Luna and the provincial government. Visit the official site.
VIA is the AI-video agency that puts these productions in front of brands. Strategy, model orchestration, and full-stack delivery.
Custom dashboards, finance tools, immersive sites, and small businesses online. Built end-to-end, deployed, and maintained.
Personal-quantified-self dashboard. Mood, sleep, focus, exercise. Vite + React + Recharts on top of Google Sheets.
Clinical-style biometric dashboard. Weight, BP, RHR, sleep stages, glucose. Vite + React stack tuned for time-series health data.
Custom-cake catalog, ordering and gallery. Designed and built for a small bakery's full online presence.
A 3D-scanned slice of the Paraná Delta you can walk through in your browser. Photogrammetry, Blender, Unity, web. An immersive love letter to a real ecosystem.
Encrypted, local-first finance manager for freelance teams. Multi-currency, debt netting, projects and payments. Open source.
Music, podcasts, science outreach. Where the technical work meets the public conversation.
Co-founded science-comms channel. Weekly videos for C5N, animated shorts for Canal Encuentro, government recognition (Ciencia por Contar 2023).
Award-winning podcast for the Argentine Society of Neuroscience. SfN Next Generation Award 2024.
Original compositions across genres. Songwriting, multi-instrument performance, and production for personal and collaborative projects.
Lead Teaching Assistant in Computational Neuroscience and Deep Learning, in both English and Spanish.
Advisor on the new primary curriculum and AI integration; educator trainer in Natural Sciences across 16 curricular areas.
For corporations, schools, ministries, and government teams. Hands-on, with custom curricula tailored per audience.
Main content creation advisor — curriculum, talks, public-facing materials for an emerging AI-safety community.
Every showcase on the site, in one wall. Click into anything.