What I Do
I work across the full ML and AI stack — building production ML systems, training
and fine-tuning models, and designing agentic AI workflows that automate real
operational problems.
Unlike most "AI engineers" who work exclusively with off-the-shelf LLM APIs, I bring
genuine ML depth: model architecture, training pipelines, classical ML where it
outperforms LLMs, and the judgment to know which approach fits the problem.
I work remotely with US-based companies on a select basis alongside my full-time work.
If you're working on something relevant, feel free to reach out on
LinkedIn.
A deep dive into MCTS — the tree-search algorithm powering AlphaGo — covering selection, expansion, simulation, and backpropagation.
AI
Algorithms
Game AI
An introduction to OpenAI Gym — the standard toolkit for benchmarking reinforcement learning algorithms across a variety of environments.
AI
Reinforcement Learning
Python
Why serverless is the right model for true auto-scaling — tying infrastructure costs directly to usage rather than provisioned capacity.
AWS
Serverless
Cloud
Hands-on experience with Upverter, a browser-based PCB design tool — from schematic capture to layout for a real hardware project.
Hardware
PCB
Engineering
Place dots on a grid and watch a tiny neural net learn the decision boundary live in your browser.
TensorFlow.js
Neural Networks
Interactive
Upload a photo and click any object — SAM instantly isolates it with a coloured mask, entirely in your browser.
Computer Vision
Transformers.js
In-Browser AI
snake-reinforcement-learning
A trained reinforcement learning agent playing Snake — demonstrating policy learning, reward shaping, and the gap between supervised and RL-based approaches to sequential decision problems.
Reinforcement Learning
Python
Local AI Chat
Chat with a small LLM running entirely in your browser — no server, no data sent anywhere.
LLMs
Transformers.js
In-Browser AI
deduz.ai
Built and launched an LLM-powered tax assistant that helps Brazilians navigate the country's complex tax filing process — reducing what typically takes hours to minutes. Covered by Exame, one of Brazil's leading business publications.
LLMs
Python
AI
Built and launched a WhatsApp-native back-in-stock notification product for Brazilian micro-boutiques that sell via Instagram and WhatsApp. Handles customer waitlists, automated alerts, and payment integration end-to-end. Currently live with real users.
WhatsApp API
Next.js
Supabase
Stripe
LLMs
pytorch-zappa-serverless
Serverless ML inference pipeline — deploying a PyTorch model to AWS Lambda with zero provisioned infrastructure, scaling automatically with demand. An example of classical ML meeting modern cloud architecture.
PyTorch
AWS Lambda
Serverless
PBandJay
An autoservice application for restaurants and bars — streamlining ordering and payments.
Mobile
Node.js
Sababa News
Collaborative
Web
