Ai Developer

Proposed embedding-based sampling methodology for whale call datasets, enhancing the efficiency of machine learning training processes. Conducted comparative analysis of various audio embedding techniques to identify sub-clusters within negative class samples, significantly improving labeling efficiency. Integrated the embedding extraction and sampling into a audio labelling tool.

Implemented stratified sampling methodologies for unlabeled sound datasets, ensuring balanced representation and minimizing potential biases. Led and coordinated a team of 10 audio labelers, establishing quality control protocols and technical guidelines for environmental sound classification. Created comprehensive documentation and training materials for audio labelers, reducing onboarding time and improving annotation accuracy.

Designed and engineered low-cost acoustic sensor solutions, optimizing energy consumption and improving accuracy of noise level measurements with signal processing techniques. Architected scalable workflows to process and ingest 2TB+ of environmental audio data into custom classification models. Fine-tuned audio classification models for environmental applications, improving detection accuracy for custom acoustic events. Programming interactive dashboards enabling stakeholders to visualize acoustic monitoring data, resulting in enhanced data accessibility and actionable insights. Authored successful research grant proposals for passive acoustic monitoring initiatives. Created automated processing pipeline for large duration audio measurements, generating classification reports, noise level assessments, and acoustic situation analytics.