Start Here

I'm Vitor Sousa, a Data Scientist at Wellhub building production-grade AI systems. My work spans LLMs, fine-tuning, RAG pipelines, and reinforcement learning — turning research into reliable applications. This site documents my projects, experiments, and the lessons learned along the way.

Recommended

If You Only Read One Thing

The Complete Guide to LLM Evaluation in Production

A comprehensive guide covering everything from automated metrics to human evaluation, A/B testing, and building robust evaluation pipelines for production LLM systems.

Read the guide

Best Articles

Deep dives into the topics I care most about — practical guides with real implementation details.

LLM Systems

LLM Evaluation Guide

Build robust evaluation pipelines for production LLM applications with automated metrics, human feedback, and A/B testing.

Read article Reinforcement Learning

Contextual Bandits Part 1

An introduction to contextual bandits for personalisation — balancing exploration and exploitation in recommendation systems.

Read article Fine-tuning

LoRA/DoRA Implementation

A practical implementation of LoRA and DoRA for efficient LLM fine-tuning, with code and experiments.

View project

What I Write About

Topics I explore regularly — from LLM engineering to production ML systems.

LLMs & Agents

Building production-ready agents with tools, memory, and planning. Fine-tuning, RLHF, and alignment techniques.

RAG Systems

Retrieval-augmented generation with smart chunking, hybrid search, reranking, and evaluation strategies.

Reinforcement Learning

Contextual bandits for personalisation, policy optimization, and adaptive recommendation systems.

MLOps & Production

Evaluation frameworks, monitoring, deployment strategies, and the infrastructure behind ML systems.