Sofia

Led AI project teams developing systems for causal inference, time series prediction, and LLM-powered data analysis. Owned end-to-end delivery from data acquisition and exploratory analysis to model design, training, evaluation, client reporting, and production-oriented implementation. - Led development of a causal failure prediction framework for wind turbines, combining sensor time series data, domain knowledge, and predictive modeling to identify early indicators of equipment failure. - Led development of an LLM-powered tool for generating, validating, and interacting with causal DAGs from datasets and metadata, with support for user feedback and on-demand causal analysis. - Supervised and evaluated interns working on AI, data science, and causal inference projects. - Contributed to research papers, technical reports, patent-related work, and internal tooling for code quality, testing, and reproducibility.

Developed multimodal image-text AI systems for radiology, with a focus on integrating visual and linguistic representations for clinical decision-support applications. - Designed and implemented models combining medical imaging with structured and unstructured textual data. - Emphasized factual accuracy, interpretability, and robustness in user-facing AI systems for high-accuracy domains. - Collaborated with an interdisciplinary team across AI, medicine, and data science.

Led the development of NLP-driven classification, knowledge extraction, and predictive modeling systems for the financial domain. - Owned end-to-end NLP model development, from problem formulation and data acquisition to model design, evaluation, deployment, and monitoring. - Built systems for text classification, information extraction, and predictive analytics. - Translated business and data problems into applied machine learning solutions.

Developed NLP tools for information extraction in the life sciences/biomedical domain. Implemented pipelines for entity recognition, relation extraction, and structured data generation from heterogeneous unstructured text, including technical documents and user-generated content.