Aluna | Data foundry for life science AI

What

Aluna is a data foundry focused on curated human datasets for life science AI. Based on the page, it appears to serve teams building biomedical and oncology AI models by providing structured dataset assets and benchmarking resources.

Its current positioning looks like a specialized data infrastructure provider rather than a general AI platform. The core workflow shown is dataset discovery and use across pathology, genomics, clinical, and benchmarking contexts, with products such as PathFoundry, Duplex, and the upcoming OncoBench supporting model development, evaluation, and multi-modal research.

Features

• Curated human datasets for life science AI — Aluna organizes human biomedical data for AI use cases, which can help research and product teams reduce time spent sourcing fragmented inputs.
• Pathology image dataset at scale — PathFoundry includes more than 2 million whole slide images across 25+ tumor types, giving teams a large pathology base for training and validating models.
• Multiple stain modalities in pathology data — PathFoundry covers H&E, IHC, and IF stains, which is useful for teams developing models that need broader histopathology representation.
• Expert-labeled slide information — Slides include diagnoses and annotations from board-certified pathologists, supporting supervised learning and more consistent downstream evaluation.
• Multi-modal biomedical dataset — Duplex combines whole genome sequencing, clinical information, and digitized pathology slides, enabling cross-modal analysis for richer life science AI workflows.
• Benchmarking suite for oncology AI — OncoBench is presented as a benchmarking product with standardized datasets, evaluation metrics, and model comparisons across cancer types and treatment modalities, though it is marked as coming soon.

Helpful Tips

• Verify dataset fit by task and modality — For life science AI, confirm whether your work needs pathology, genomics, clinical data, or benchmarking, since each Aluna dataset appears aimed at a different stage of model development.
• Review annotation scope carefully — Board-certified pathologist annotations are valuable, but buyers should still assess label definitions, coverage, and consistency before committing to a training pipeline.
• Plan for multi-modal data alignment — If using a resource like Duplex, make sure your team can handle data linkage across genomic, clinical, and image records, as this often drives implementation complexity.
• Separate benchmark use from training use — A benchmarking suite such as OncoBench is most useful when kept distinct from model training data to avoid evaluation leakage.
• Ask about access details not shown on the page — The page does not specify licensing, delivery format, update cadence, or technical documentation, so those factors should be clarified during evaluation.