At Data4Sales, I designed and deployed a data-analysis agent used by more than 240 companies. It answers business questions by running SQL in parallel and analyzing results in a Python sandbox, then delivers reports or campaign audiences.
I also replaced collaborative filtering with a multi-source retrieval and LightGBM LambdaRank recommender. It improved offline Hit@5 by 26%, recommended to customers without history, and reduced peak inference memory for the largest company from 19.9 GB to 4.9 GB without changing the output.
At AccelOne, I secured AI systems for a leading LATAM retailer with guardrails, environment controls, and IAM permissions for agents and models on AWS Bedrock. I also worked on production ML deployments and computer vision for a baseball app, including pose detection and on-device object tracking.
Across Poderify and Adava, I delivered RAG, NLP, recommendation, and forecasting systems, with MLflow-based tracking, deployment pipelines, and drift monitoring. Outside work, I built AutomationTrading Suite, an open-source Python platform for quantitative research and algorithmic trading.
