AI Engineering student at Universidad de San Andrés (UdeSA), Buenos Aires. I build agentic systems end to end — from models implemented from scratch in NumPy to production backends powered by LLMs.
🌐 luisembon.vercel.app · ✉️ luisembonstrizzi@gmail.com
- 🏆 Winner — Anthropic × Kaszek × Digital House hackathon (Apr 2026): 1st place in e-commerce and 2nd overall among 55 teams. Now building Prompty, an AI listing optimizer for Mercado Libre (FastAPI · DSPy · Claude · Supabase) — in public beta.
- 🤖 Building an agentic dispute-resolution platform for fintech: Claude agents orchestrated with n8n over Postgres/MongoDB, grounded with RAG.
- 🚗 Multimodal trajectory prediction with a CVAE on the rounD roundabout dataset — final project for I302, presented as a poster at UdeSA AI Fest 2026.
- 🩺 Applied computer vision for pediatric dermatology, in collaboration with a physician from Hospital Italiano de Buenos Aires.
A series of projects built to master the fundamentals — every model implemented from scratch in NumPy:
| Project | What's inside |
|---|---|
| Housing regression | Pseudo-inverse & gradient descent, L1/L2, stratified splits, segmented models |
| Student risk classification | Logistic regression, LDA, Random Forest; group/temporal splits, rebalancing |
| Kuzushiji neural nets | MLP with manual backprop, Adam, scheduling; PyTorch + augmentation on Kuzushiji-49 |
| Fashion-MNIST unsupervised | PCA, k-Means, GMM, t-SNE from scratch; autoencoder & β-VAE in PyTorch |
Portfolio source: portfolio — Next.js 16 · Tailwind v4 · shadcn/ui, deployed on Vercel.