Adriel Martins

Built multi-agent RAG and Computer Vision systems that cut customer-support response times by 50% and accelerated logistics analyst actions by 20%, delivering 4x analyst throughput improvements in delivery monitoring.

Led production LLM and agent architectures delivering 95% accuracy in enterprise video retrieval and engineered scalable RAG/vector DB pipelines with automated ingestion for real-time search and 5x speedups in sports analytics.

Engineered training and deployment pipelines that accelerated model development by 75% and reduced equipment downtime by 20% via outlier detection and data monitoring in condition monitoring tools.

Built end-to-end data aggregation and RAG systems that increased revenue by 30% and improved classification accuracy by 5%, implementing vector DB retrievals and GPU-accelerated inference to process brand monitoring data 3x faster.

Delivered NLP topic-tagging and recommendation systems that sped chatbot development by 50% and raised call acceptance projections by 10%, and doubled effectiveness of airport-targeted marketing via geospatial analysis.

Developed parallelized ETL data products and pipelines that improved ingestion performance by at least 30% and enriched client datasets for market-research insights.

Built queueing-theory models and GIS-based pipelines to forecast navigational processes and provided interactive dashboards for harbor operations analysis.

Led development of interactive dashboards and multivariate analyses (PCA, GARCH) that improved RECIPREV decision flow by 50% and proposed new metrics to enhance financial decision-making.

Activities and societies: Research and dissertation on Graph Neural Architecture Optimization (GraphNAO); coursework in advanced machine learning and graph neural networks.

Pursuing an MSc in Computer Science with a dissertation (GraphNAO) on automated GNN architecture optimization to improve retrieval and reasoning in graph-based RAG systems; expected completion in 2025.

Activities and societies: Undergraduate thesis research; published work in time series analysis and information theory.

Completed a Bachelor's degree in Statistics with a thesis developing a confidence interval estimation method for Bandt and Pompe analysis applied to 59 years of quarterly GDP data across 25 OECD countries.

Activities and societies: Project-based coursework in computer vision, including practical model implementation and evaluation.

Completed the Computer Vision Nanodegree focusing on practical computer vision techniques and model deployment.