AI Engineer · AI Researcher
Leonardo Arazo
> agent.observe(fresco) → 2 hands · 0.4 cm apart
> agent.reason() → contact is imminent
> agent.act("create") ✓
Where human intuition
touches machine intelligence.
I — Manifesto
On engineering & research
Gabriel
the messenger · inked small, on me
i.
Engineering
Research taken to production: per-user knowledge graphs, and agent memory that cut token costs by 79%.
ii.
Research
Deep learning for rain at INPE, neurosymbolic AI, and a direct PhD.
iii.
Publication
KDMiLe / BRACIS 2026 with FespAI, and image compression at SBPO.
The machine learns as the hand does: by repetition.
An alphabet
of me
Renaissance workshops marked their work with symbols. These are mine: what I research, what I build, what I wear and what is inked on me.
- 01Rain forecasting · INPE
Pluvia
the rain to come
- 02Remote sensing
Satelles
the eye in orbit
- 03RAG · FespAI
Codex
the book that answers
- 04AI agents
Agens
the one who acts
- 05Knowledge graphs
Graphum
the map of knowledge
- 06Neurosymbolic AI
Mens et Ratio
intuition meets logic
- 07Neural networks
Homo Machina
the measure of models
- 08SVD & compression
Sigma
the decomposition
- 09Computer vision
Oculus
the eye that sees
- 10Optimization
Scala
the descent
- 11My tattoo
Gabriel
the messenger, on my skin
- 12Streetwear
Catena
the chain, always on
Scattered data, waiting for a form.
Homo Vitruvianus, now in streetwear. The measure of man, c. 1490.
Move your cursor through the ink
The measure
of machines
The chronicle
Every chapter is printed in ink dots, the way an image looks to a machine before it understands it.
I · Present
deco
Senior AI Engineer
deco is an open-source platform for AI agents: a control plane to connect tools, hire agents and track every token, plus an agentic CMS for storefronts.
- AI agents
- MCP
- Agentic CMS
II · Present
AMAWeb
Software Researcher
Web accessibility evaluation and monitoring at UNIFESP: free tools that automatically evaluate web pages, find accessibility barriers and recommend fixes, plus an observatory of accessibility indicators.
- Web accessibility
- Automated evaluation
- Observatory
III · Present
INPE
AI Researcher · Rain forecasting
Precipitation forecasting with deep learning at the National Institute for Space Research. Second place nationwide in the INPE rain forecasting competition, with team Rain-NP-Hard.
- Deep learning
- Rain forecasting
- 2nd place · Rain-NP-Hard
23°12′S 45°51′W · INPE
radar · precipitation · t+1h
IV · Founding team
Isla
Founding AI Engineer
Founding team of isla, a San Francisco startup that turns LinkedIn into booked sales meetings. Built per-user knowledge graphs with a neurosymbolic pipeline for personalized content ideas, and memory management for AI agents that cut token costs by 79%.
- Knowledge graphs
- Neurosymbolic
- AI agents
- −79% tokens
V · Engineering
Zazos
AI Engineer
Developing RAG systems and implementing scalable AI solutions in production.
- RAG
- LLMs
- Production
VI · 2023 – 2025
UNIFESP
B.Sc. Computer Science · AI Researcher
Creator of FespAI, the neurosymbolic academic assistant of UNIFESP, published at KDMiLe / BRACIS 2026 and first place at CTIC · WebMedia 2026. Research on image compression with SVD and Autoencoders, presented at SBPO. Teaching assistant in Analytic Geometry.
- FespAI
- KDMiLe / BRACIS
- SBPO
- Teaching assistant
The laurels
The problem is rarely the model. Understand the problem and the data first; the results follow.
- 01FespAI1st
CTIC · WebMedia 2026
First place at the SBC scientific initiation contest, plus an honourable mention at the Luiz Fernando Soares Award for social impact.
- 02INPE2nd
National rain forecasting competition
Second place nationwide in INPE's precipitation forecasting competition, with team Rain-NP-Hard.
- 03FespAI68.4%
Error rate, down from 68.4%
The neurosymbolic pipeline of FespAI, running only on local open models, at R$ 0 in API costs.
- 04isla−0%
Token costs in production
Memory management strategies for AI agents, alongside per-user knowledge graphs feeding a neurosymbolic content pipeline.
- 05SentiNela0%
Precision on symptom extraction
SentiNela, an AI syndromic surveillance system selected for Science & Business.
- 06PublicationKDMiLe
Paper at KDMiLe / BRACIS 2026
The neurosymbolic academic assistant behind FespAI, published.
- 07FespAIUSP · UFSCar
Adopted beyond UNIFESP
Professors at USP and UFSCar are adapting FespAI for their own universities.
- 08UNIFESPFase Zero
SBC Programming Marathon
Honourable mention, and a standout teaching assistant in Analytic Geometry.
Il custode
Scroll ↓ · or poke him
Let's paint
something new
You don't have to.