At Kunzapp, I own the ML/AI stack and data infrastructure. I shipped a RAG pipeline on AWS Bedrock that reduced lookup time by 75% and support tickets by 40%, along with an evaluation framework for groundedness, cost per query, and human review.
I built an LLM invoice classifier on AWS Bedrock, improving accuracy from 60% to 95% across about 2,000 documents a day, and developed LangGraph agents that query company databases through MCP. I served models with FastAPI on AWS, reducing inference latency by about 90%, and manage four production models with MLflow.
Before Kunzapp, at Cobalto Inmuebles, I owned ML DevOps for three production models and built cohort, survival, and churn models to guide retention strategy. As a freelance data scientist, I built invoice fraud-detection and demand-forecasting systems, including an airline candy-sales forecast paired with an optimization model.
My first-author paper, “Metro Feeder Route Optimization,” was published in Springer CCIS vol. 2554. I work across PyTorch, TensorFlow/Keras, and Scikit-learn, and founded a biweekly AI/ML workshop at Kunzapp.
