Rio de Janeiro, Rio de Janeiro, Brazil
I’m an AI Engineer Specialist with more than eight years of experience applying machine learning and natural language processing to solve real-world business challenges. My background spans industries like oil & gas, finance, legal tech and e-commerce, where I’ve delivered solutions ranging from form auto-filling with NLP and contract clause extraction to sentiment analysis and customer retention modeling. At Digibee, my focus is on bringing NLP and Generative AI into product-ready pipelines always with an emphasis on scalability, reliability, and user value. I’m passionate about bridging the gap between experimentation and production, ensuring that every AI feature we deliver is practical, explainable, and impactful for our customers.
Worked on the improvement and maintenance of production-ready machine learning models for Ambev’s recommendation system on the Bees App, a large-scale B2B e-commerce platform. Worked closely with cross-functional teams to align model behavior with domain-specific rules and commercial objectives, all within a scalable, distributed PySpark-based environment. • Implemented a post-processing calibration model using isotonic regression, improving score interpretability for business stakeholders. • Contributed to the optimization of business rule weights using Bayesian optimization, enhancing alignment with strategic goals. • Designed and executed A/B tests to validate the impact of business rules on recommendation performance and user engagement in production. Tech Stack: Python · PySpark · Azure Databricks · Airflow · Github · Bayesian Optimization · Isotonic Regression · A/B Testing · Statistical Modeling · Recommendation Systems
Worked on the development and optimization of machine learning models designed to extract M&A-related clauses from legal contracts for three of São Paulo’s largest law firms. Focused on improving performance, interpretability, and cost-efficiency of the legal NLP pipeline. • Tuned an existing classification model using Bayesian optimization, enhancing its effectiveness on domain-specific legal texts. • Explored and evaluated alternatives to OpenAI embeddings, such as Falcon models, to reduce vendor dependency and control operational costs. • Collaborated with legal experts to validate clause extraction outputs and improve model alignment with real-world legal use cases. • Worked in a JupyterLab-based environment on AWS, contributing to experimentation and delivery in a cloud-native ML workflow. Tech Stack: Python · AWS · Bayesian Optimization · Hugging Face · Falcon Models · OpenAI
Led ML initiatives focused on equipment failure prevention and NLP-driven process automation for industrial IoT operations in the oil & gas sector. Delivered business-critical machine learning systems that significantly improved both operational efficiency and model interpretability. • Developed an NLP-based automation tool for form pre-filling and report processing, saving hundreds of hours in manual review and accelerating decision workflows. • Replaced a legacy Deep Learning model with a linear model enhanced by business-driven features and interpretable encoding, raising anomaly detection precision from 5% to 70%, enabling the successful detection of five pre-failure events—preventing unplanned equipment downtime and reducing operational risk. • Designed a trend detection framework using Continuous Trend Labeling combined with Bayesian optimization, emulating a reinforcement learning-style feedback loop to detect abnormal thermal patterns in industrial sensor data. • Worked in a cloud-based ML environment using Azure Databricks, alongside in-house tooling built on top of MLflow and GitFlow, adapted for internal pipeline, packaging, and experiment tracking workflows. Tech Stack: Python · Azure Databricks · Time Series Modeling · Bayesian Optimization · BERT · ML Pipelines · Hugging Face
Developed and deployed machine learning models for financial trading, specializing in trend-following strategies applied to time series data. Collaborated closely with quantitative research teams to test and evaluate live models that enhanced portfolio performance. • Developed and deployed two trend-following algorithms, each delivering over 15% annualized returns in live trading, outperforming benchmark strategies and directly contributing to investment results. • Built a deep learning model for trend forecasting using supervised signal labeling and custom feature engineering, leading to one of the firm’s top-performing strategies. • Designed and backtested a rule-based trading system, enhanced with Bayesian optimization for parameter tuning, generating robust and stable trend signals (algorithm details under NDA). • Prototyped a financial NLP pipeline using Hugging Face’s BERT to summarize real-time market news, aiming to enrich quant models with textual insights. Tech Stack: Python · Tensorflow Keras · Hugging Face · BERT · Bayesian Optimization · Time Series Modeling · Financial NLP · Algorithmic Trading