Rodolfo Teles - Machine Learning Engineer

Rodolfo Teles

Machine Learning Engineer | MLOps Specialist

With 6 years of robust experience in designing and maintaining MLOps pipelines. Expertise in end-to-end machine learning projects, ETLs/ELTs pipelines, and software development for global leaders in digital transformation.

Intensifying studies to achieve greater proficiency in Retrieval-Augmented Generation (RAG) and LLMOps, while currently deepening expertise in diffusion-based generative models.

Professional Achievements

MLOps Pipelines

In Globo’s Data and Recommendation Systems teams, managed the entire lifecycle of machine learning projects, from acquiring and cleaning large volumes of data using BigQuery, PySpark, and SQL to deploying them into production and presenting results for business team.

Architected MLOps pipelines in Vertex AI, Dataproc, Pub/Sub, and Cloud Functions that supported an entity resolution model capable of deduplicating millions of GloboID records with reproducible parameters and interpretable metrics for the business.

In addition, managed the entire lifecycle of machine learning projects, from data acquisition in databases to presenting results in Tableau, and orchestrated the pilot implementation using AWS services of a recommender system for a multinational client, resulting in a 35 % increase in net sales of suggested products.

Generative AI & RAG Search

Delivered advanced AI/ML solutions leveraging Google Cloud Platform (GCP) services such as Vertex AI, Vector Search, Cloud Run, and BigQuery. Designed and developed Retrieval-Augmented Generation (RAG) pipelines to enhance information retrieval using Generative AI and Large Language Models (LLMs).

Led the creation of a semantic search engine optimized for educational content, integrating RAG and LLMs, and deployed the solution with Docker and Google Kubernetes Engine (GKE), featuring a Gradio-based interface to allow user engagement.

Contributed to the design and implementation of a multimodal search framework, ensuring high software quality through robust formatting, linting, and automated pipelines with GitHub Actions, while utilizing Vertex AI embeddings, BigQuery, and FastAPI for efficient CRUD operations. Participated in deploying Llama 3.1 (405B parameters) on Google Compute Engine, demonstrating expertise in scaling LLM solutions for production environments.

Generative AI with Diffusion Models (Text-to-Music)

Contributed to the development of generative AI models applied to music creation, using diffusion models to generate spectrograms decoded in high-fidelity audio. Designed and fine-tuned workflows that enabled real-time music generation, combining stable diffusion techniques with audio signal processing.

Fine-tuned diffusion models specifically for spectrograms, improving audio quality and music composition. Collaborated with an interdisciplinary team of musicians, engineers, and business professionals to question the limitations of a text-to-music model and conduct several experiments to validate hypothesis related to outputs limitations (audio quality, duration and composition).

Education

Mechatronics Engineering

UFRN (Ranked 31st among Latin American universities by Webometrics)

MLOps Specialization

DeepLearning.AI

Machine Learning Engineering

Udacity Nanodegree

Core Skills

  • Generative AI & LLM Engineering
    • LLM/LangChain
    • RAG Systems
    • Diffusion
  • MLOps & LLMOps
    • End-to-End CI/CD
    • Model Fine-Tuning
    • Monitoring & Guardrails
  • Cloud
    • SageMaker
    • AWS
    • GCP
    • Vertex AI