About this role
Build foundational substrate for the AI for Science Gym including data ingestion pipelines, modular model interfaces, LLM- and RAG-assisted dataset and workflow interfaces, and components that scale from single-GPU to multi-node HPC with containerised, scheduler-friendly deployment on NUS HPC.
Skills for this role
PythonPyTorchData pipelinesMetadata schemasProvenance trackingFormat normalizationHDF5TIFFOME-TIFFZarrModular Python APIsTokenization for non-text modalitiesRepresentation learningSelf-supervised learningFoundation modelsLLMsRetrieval-augmented generation (RAG)EmbeddingsVector storesNotebook UIsDashboardsViewer applicationsAnnotation applicationsGPU computeMulti-node HPC scalingContainerisationDockerApptainerSlurmDistributed training''Reproducible ML workflows''Testing''Software/documentat