Rubens Zimbres, Ph.D.

2024-09 → now · Zendata · San Francisco, remote

Machine Learning Engineer / AI Security

Code and model privacy, PII protection, RAG and LLM red teaming at a SOC 2 certified company. I created LLM-Scan, a product that evaluates LLMs with thousands of crafted prompts across Security, Legal, Brand and General risk. Deployed on AWS Batch, SageMaker, Bedrock, Azure OpenAI, Google Gemini, Cloud Run GPUs and Ollama.

I also work on the security and privacy of multi-agent systems built with CrewAI and LangChain. The CEO of Zendata and I presented this research at BlackHat Europe 2024.

2024-05 → now · Google for Startups

Mentor, Google for Startups Accelerator Brazil

2022-05 → now · Google Developers

Google Developer Expert (Machine Learning, Google Cloud) and Google Cloud Innovator

I build solutions that combine Google Cloud services with Machine Learning, deploy them, and write about the experience for the Google Developer Experts and Google Cloud blogs. As a Cloud Innovator I test new Google Cloud technologies.

2019-04 → 2023-06 · Intellimetri

Senior Data Scientist / Machine Learning Engineer

Intellimetri is a spin-off of Vecto Mobile and a Google Cloud Partner. I was responsible for ML algorithms, pipelines, MLOps, end-to-end architecture and the cost of our Google Cloud infrastructure, with a focus on NLP and recommendation systems.

• Two-stage recommender for a Virtual Career Center, trained and deployed daily to Vertex AI endpoints through Kubeflow pipelines. Delivered to a municipality in Colombia.
• Highly scalable retrieval model for a government agency in Mexico, on Vertex AI Matching Engine.
• Dialogflow chatbot to search Brazilian laws by embedding similarity (Two Towers), served from Cloud Run.
• Computer vision for supermarkets: counting milk cans at the cashier to prevent fraud.
• Revenue forecasting for a 130-unit hotel chain with Temporal Convolutional Networks: 91.5% average accuracy over a 3-month horizon.
• Speech analytics and automated monitoring of call center voice calls.
• Algorithmic swing trading on BOVESPA with Compute Engine and cron jobs.

This work produced two NLP patents (2020 and 2023). I also mentored an Applied Mathematics undergraduate at the University of São Paulo whose final thesis scored 10/10.

2017-10 → 2019-04 · Vecto Mobile

Data Scientist

Research and development of Machine Learning and Deep Learning products around speech recognition and prediction. Our NLP work for call centers took us into Google Cloud for Startups, and later to Google Cloud Partner status.

We deployed a virtual career solution in 6 Latin American countries using face recognition, sentiment analysis, speech analytics and NLP: up to 73% less screening time, up to 78% less time to fill vacancies, over 85% accuracy in choosing candidates. By December 2022 more than 365,000 people had used it.

• Predictive maintenance from point-of-sale telemetry.
• Scientific coffee roasting with IoT sensors.
• Voice fraud detection with speech-to-text, word embeddings and gradient boosting.
• End-to-end streaming IoT telemetry on AWS (IoT Core, Kinesis, Firehose, DynamoDB).

2019-03 · Google and UOL Diveo

Lecturer

Presented our business case (NLP, speech recognition and deep learning on Google Cloud) to managers and directors of large Brazilian retailers.

2018-02 · Goiás University

Instructor, Deep Learning Brazil Summer School

Lectured on Generative Adversarial Networks and ran a hands-on Keras workshop.

2010-10 → 2013-07 · Doux Dermatology

Business Intelligence Analyst

Doubled profit in the first year through digital marketing, market research and customer focus. After 2.5 years: net profit +239%, ROI +206%, ROE +95%, EBITDA +98.5%.

2005-02 → 2009-12 · DR3 Consulting

Consultant

My own consulting firm, specialized in strategic management and market research analysis.

1995 → 1997 · Brazilian Air Force

1st Lieutenant Dentist, Head of the Health Sector

Responsible for military training, facilities security and health-sector logistics. Volunteered on the mission that rescued Brazilian tourists from St Maarten after Hurricane Luis.

2006 → 2009 · Mackenzie Presbyterian University

Dual Doctorate, Business Administration and Electrical Engineering

Agent-based modeling in management: searching for a cellular automaton rule able to simulate human interactions in business services. Full 4-year merit scholarship, with additional training sponsored by Wolfram Research. Resulted in a patent and a paper in Elsevier's Electronic Notes in Theoretical Computer Science.

2006 · Brown University

Wolfram Research Summer School

Studied neighborhood effects in the cellular automaton model of my doctorate thesis, using Mathematica.

2004 → 2005 · Mackenzie Presbyterian University

Dual Master, Business Administration and Electrical Engineering

Cellular automata simulating human rationality and interactions as social networks. Dissertation: "Dynamics of formation and evolution of business networks in Dentistry". Full 2-year merit scholarship.

2005 · University of São Paulo, Economics

Complexity, Game Theory and computational modeling

Built a model of human interactions based on the Standing Ovation Problem of Miller and Page.

1992 · University of São Paulo

Dentistry