Daniel Manzato

About Daniel Manzato

With more than 25 years of experience in software development, I currently work as a Senior Software Development Engineer at AWS, where I have consistently delivered resilient, scalable, and high-performance solutions within the Artificial Intelligence and Machine Learning organization. My role involves designing, developing, and maintaining distributed services, with a strong emphasis on security and operational excellence. My work in machine learning and generative AI leverages AWS services such as Bedrock, SageMaker, and OpenSearch to create innovative solutions.I have extensive experience with programming languages such as Java, C/C++, and Python.I am committed to delivering customer-centric solutions that meet the highest standards of quality and reliability.

Experience

1. Senior Software Development Engineer

   Amazon Web Services (AWS)

   Aug 2019 — Present · Seattle, WA, US

   AWS Q Business (2024 - present): responsible for image extraction from different media types to leverage Model Vision in GenAI-based agents. Developed content extraction service to securely open files uploaded by customers and return extracted media such as text and images to be used in LLMs. Proposed approach for optimally resizing images before sending them to LLMs in binary format; this approach was leveraged later by other teams in the org. Currently working with 3rd party app integrations with LLMs.AWS Sensor Data Search (20••••24): partnered with Science team to design and develop a similarity search service for sensor data based on embeddings indexed in a vector database. Text and image-based queries were supported enabling customers to search for specific objects, scenarios and situations in extensive data such as videos coming from Autonomous Vehicle cameras. Accountable for a index restructuring proposal based on OpenSearch\'s parent-child documents which achieved a perfomance improvement of 76% on search latencies and a reduction of 41% on embedding storage by diminishing redundancy. The proposed approach ended up being leveraged by other teams using OpenSearch internally.Amazon Fraud Detector (20••••23): supported GA launch and product evolution by working in the Auto ML team, responsible for model retraining pipeline and partnering with Science team to support MDLC. Contributed with design, development, testing, ops improvements of different components such as a training dataset generator, used to leverage customer events pre-ingested into partitions in a data lake to produce datasets, and automated generation of ML Docker containers, used as building blocks in inference and training pipelines. Additionally, I spearheaded the design and development of a new Model Type for detecting and preventing Account Takeover attacks, leveraging negative-sampling synthetic data generation to support supervised learning algorithms with limited labels.

Education

• Universidade Estadual de Campinas

  Master of Science (MSc), Computer Science

  2006

• Universidade Estadual de Campinas

  Doctor of Philosophy (PhD), Computer Science

  2015

• USP - Universidade de São Paulo

  Bachelor of Science (BSc), Computer Science

  1998