Andres Olguin

Implemented an end-to-end Lakehouse in Databricks using Bronze, Silver and Gold layers over a real sales dataset (~15M records). Designed and implemented a star-schema model with sales fact tables and dimension tables (time, store, product), optimizing analytical queries. Used PySpark and SQL for data ingestion, cleaning and transformation, type handling, field normalization and quality checks. Worked with Delta Lake (ACID, time travel, partitioning), Unity Catalog, Volumes and Jobs to orchestrate the recurring execution of notebooks. Documented the architecture, naming conventions and process flows for the technical presentation of the project.

Designed and evaluated multi-turn conversations to train and refine a Large Language Model (LLM) for real-world assistant behavior; simulated user-assistant interactions involving reasoning, tool usage, and multilingual responses; ensured logical flow, contextual accuracy, and adherence to high-quality labeling and reasoning standards defined in the Penguin ACPS i18n framework; collaborated with reviewers and leads through Slack and Jibble; focused on improving AI tool-call consistency, error recovery, and natural dialogue generation across multiple domains (Calendar, Gmail, Maps, Web Search, etc.).

Led a 7-person development team to design and build the complete tech stack for a sports-oriented fintech startup. Responsibilities included task planning, delivery reviews, Jira tracking, coordinating with mentors and the CEO, documenting processes, defining best practices, infrastructure support, provider interviewing, solution analysis, and fostering team culture with a manifesto and improved communication.