Sebastián Castaño

Conception and development of a framework for knowledge management and discovery in machine learning teams. The product is framed as a discovery engine for machine learning research, tackling the problem of information overflow like more than 100 machine learning papers uploaded daily to arXiv.

Key Activities
• Implemented a PoC system consisting of a discovery engine for machine learning research using large language models.
• Designed and deployed cloud infrastructure serving the discovery engine.
• Created a business model and go-to-market strategy.
• Conducted user discovery and development interviews with more than 50 interviewees.

Development, implementation, and deployment of a marketing mix model based on Bayesian statistical modeling for the company's global operations.

Key Activities
• Implemented a learning model based on state-of-the-art research papers.
• Customized the model based on specific properties of the data available.
• Deployed the model to the cloud to be used by the MLOps team.
• Improved the model iteratively using feedback from the business unit.
• Presented results to several non-technical stakeholders in the business unit.

The first machine learning-based adaptive deep brain stimulation system implemented in freely moving essential tremor patients. This project was executed in cooperation with colleagues at the University of Washington.

Key Activities
• Established the cooperation between our research lab at the University of Freiburg (brain state decoding lab) and the University of Washington (biorobotics lab).
• Conceived and developed the underlying machine learning, control, and digital signal processing methods.
• Deployed the algorithms on a host PC and the embedded system of the patients' neurostimulators.
• Executed the data collection experiments.
• Performed offline analysis of the collected data.
• Wrote and edited the final manuscript for a peer-reviewed publication.

Descriptive and prescriptive analysis of injury data for a Bundesliga soccer team. After I analyzed the data and presented the results to all the stakeholders, including non-technical personnel, the project became part of the team's research department to investigate injury prevention and talent development.

Key Activities
• Prepared and cleaned the data from several databases.
• Conducted descriptive and prescriptive analyses of the data.
• Presented the results to all stakeholders, including non-technical personnel.

Handover of a data management system that a large European bank used for the warehousing and analysis of credit risk data.

Key Activities
• Configured and deployed UAT and production environments.
• Implemented a CI/CD pipeline for the back and front end.

A novel supervised ML approach for decoding the intensity of Parkinson's disease symptoms from brain signals. We collected data from seven patients undergoing deep brain stimulation therapy and showed that our novel ML approach improved the decoding performance of the symptoms.

Key Activities
• Designed and executed the data collection experiments.
• Preprocessed data and performed exploratory data analysis.
• Conceived and implemented the novel ML method.
• Validated the novel ML method with the collected data and a benchmark against state-of-the-art models, including deep convolutional neural networks.
• Applied AutoML for hyperparameter optimization of all considered models.
• Wrote and edited the final manuscript published in a peer-reviewed journal.

The novel framework developed in this project allows for an objective evaluation and benchmarking of novel ML algorithms to analyze neurological data.

We tackled the following challenges:
• Scarcity of data available when using data-driven methods in the analysis of brain signals.
• Unreliability of the available labels
• High level of noise in the raw signals.

Key Activities
• Conceived the idea.
• Executed the data analysis.
• Wrote the scientific manuscript published in a peer-reviewed journal.

Served as a tech lead on the global data science and data analytics team and played a key role in the conception, implementation, and deployment of ML-based optimization of advertising strategies for global operations.

A proof of concept of a fully automated data-driven A&R for a music label with the objective of developing music artists at scale. The resulting machine learning system comprises the following subsystems: artist breakthrough prediction, future streaming revenue forecasting, and artist development strategy optimization.

The proof of concept was developed in the context of Antler, a pre-idea startup accelerator. With the POC and the developed business plan, we raised $100,000 in pre-seed funding.