Buenos Aires, Argentina

Luis Embon Strizzi

I build AI agentsend to end — from first principles to production.

Luis Embon Strizzi
  • Hackathon winner

    1st place (e-commerce) & 2nd overall of 55 teams — Anthropic × Kaszek × Digital House, Apr 2026

  • AI Engineering @ UdeSA

    Universidad de San Andrés — Ingeniería en Inteligencia Artificial

  • From scratch to production

    NumPy-level fundamentals + LLM agents, RAG and full-stack apps

stack.yaml
read-only

# tools I ship with

ml_research:

  • -Python
  • -PyTorch
  • -NumPy

llm_agents:

  • -Claude API
  • -DSPy
  • -RAG
  • -n8n

backend:

  • -FastAPI
  • -Supabase
  • -PostgreSQL
  • -MongoDB

web_infra:

  • -Next.js
  • -TypeScript
  • -Tailwind CSS
  • -Docker
  • -GitHub Actions

Projects

A latent space of everything I've built

Every project embedded as a point — related work clusters together. Drag your mouse to look around, then click a cluster to fly into it.

projection: t-SNE(work) → ℝ³ · perplexity 30

ml-research · training run

Six checkpoints of decreasing loss

epoch 0/40val_loss 1.794 ↓
0.250.751.251.75010203040val_lossepochs (2025 → 2026)ckpt-01ckpt-02ckpt-03ckpt-04ckpt-05ckpt-06
scroll to train

llm-agents · flagship

Prompty.

Prompty — AI listing optimizer for Mercado Libre
Hackathon winner — Anthropic × Kaszek × Digital House ↓

AI listing optimizer for Mercado Libre

Listings optimized with AI and real marketplace data: Prompty scans a listing, audits it against the top sellers of its category, and generates optimized titles, descriptions and attributes in seconds. I own the backend: FastAPI, DSPy pipelines compiled with MIPROv2 on Claude, Supabase and the Mercado Libre OAuth integration. Built with Ciro Vilmer & team — now in public beta.

FastAPIDSPyClaudeSupabaseNext.jsMIPROv2
Try the live betabuilt with Ciro Vilmer & team

llm-agents · session

Chat with my work

The agents side, told the way I build it: a tool call fires, evidence comes back, and only then the answer streams. Scripted from real work — the streaming is the only simulation here.

session.log
status
idle
transcript
scripted · real work
tools
DSPy · RAG · n8n
tokens_out
0/370
turns
1/4
luis — agent session
tool-use: on
Ask about my work… ▍scripted demo

// full index

  • 01
    Prompty
  • 02
    Agentic dispute platform (fintech)
    Code private
  • 03
    Paper Scout
  • 04
    Multimodal trajectory prediction
  • 05
    AI in pediatric dermatology
  • 06
    Housing regression
  • 07
    Student risk classification
  • 08
    Kuzushiji neural nets
  • 09
    Fashion-MNIST unsupervised

Awards

Winner · April 2026 · Buenos Aires

Anthropic × Kaszek × Digital House Hackathon

We built Prompty in 8 hours: an AI listing optimizer for Mercado Libre. We won the e-commerce category and placed 2nd overall among 55 teams, presenting live on the final stage in front of 200+ people.

e-commerce category
#0

e-commerce category

overall · 55 teams
#0

overall · 55 teams

in Claude API credits
$0k

in Claude API credits

cash prize
$0k

cash prize

Judged by leaders fromAnthropicKaszekMercado LibreTapiDigital House

About

I'm a third-year AI Engineering student at Universidad de San Andrés (UdeSA). I like understanding things from first principles — most of my coursework implements the models from scratch in NumPy before reaching for a framework — and then shipping them as real products.

In parallel with my coursework I'm always building projects of my own: LLM agents and production backends on Claude, applied ML in fintech and medical imaging, and whatever the next idea demands — everything on this site comes from that habit.

Get in touch

Open to internships, research collaborations and interesting agentic-systems problems.

Email meGitHub