Maurício – Python, AWS, LLM

It's a data workflow that matches verbatim with existing engineering/meta-function issues.

• Built the full POC to validate the LLM-based solution.
• Implemented function calling on both client and server within the internal GenAI platform.
• Designed and optimized a RAG pipeline with an in-memory database.
• Applied prompt engineering techniques to improve model outputs.
• Developed data connectivity for the MVP, integrating external APIs and internal systems.
• Defined and delivered data engineering needs for the MVP.
• Designed and specified API requirements to support the MVP.

It's a data workflow that processes Warranty Claims verbatim from dealers and provides a Dashboard to Quality pre-analysts with the main component-defect pairs.

• Developed comprehensive regression tests (unit & integration).
• Designed the end-to-end solution using sub-sample clustering and UMAP-based dimension reduction for scalability.
• Identified and resolved sparse data issues with UMAP.
• Implemented 30% of the full workflow in Databricks.
• Provided technical guidance to both Data Engineers and Data Scientists.
• Managed integration with Palantir Foundry, handling both inputs and outputs.
• Led all LLM-related development, being the project’s sole GenAI expert.

It's a GenAI Gateway for compliant LLM access.

• Gathered and managed feature requests, translating real use cases into product development.
• Built a unified integration with Bedrock and OpenAI models.
• Implemented function calling, LLM parametrization, and prompt engineering.
• Transitioned the system to a multi-modality approach leveraging new LLMs.
• Ensured reliability by writing unit tests, debugging issues, and resolving application errors.
• Produced documentation and delivered a live platform demo to 600+ employees at the company.
• Provided ongoing client support throughout the project.

It's an algorithm to find hierarchical patterns in proteins.

• Designed, developed, and deployed two solutions from scratch: a heuristic search and a deep learning-based approach.
• Built a Streamlit front-end for data ingestion and visualization.
• Implemented a heuristic search algorithm leveraging amino acid count distributions.
• Developed a Deep Variational Autoencoder trained on synthetic data to learn protein embeddings.

It's a venture capital analyst agent that scrapes the Internet and provides a complete report about a certain startup.

• Developed the full solution leveraging LangChain and LlamaIndex.
• Built a Streamlit front end enabling interactive rule additions.
• Containerized the application with Docker Compose using a Selenium image.
• Integrated the OpenAI API for LLM capabilities.
• Deployed the solution on AWS for scalability and accessibility.