MullionAI
For AI labs
Data your model hasn't seen.
Licensed, non-public media, pharma and health datasets for pre-training, fine-tuning and evaluation. Every set comes with clear rights and full provenance.
Media & content
Video, audio and text.
Professional content from studios, broadcasters, labels and publishers, with metadata included.
Scripted film & TV
Professionally shot drama and comedy, with scene, shot and dialogue metadata.
Request dataset →Documentary & archival
Decades of dated, captioned footage covering real places, people and events.
Request dataset →Sports & multi-camera
Synchronised angles of live events, for motion, physics and world models.
Request dataset →First-person footage
POV recordings of people doing real tasks, for robotics and embodied AI.
Request dataset →Music & audio
Licensed recordings, stems, podcasts and speech, with rights cleared at track level.
Request dataset →Books & editorial
Edited long-form text from publishers, for reasoning, writing and evaluation.
Request dataset →Pharma R&D
Science that never got published.
Internal research data from pharma companies, de-identified and licensed for scientific and medical AI.
Clinical trial records
Protocols, outcomes and analyses, including terminated and unpublished trials.
Request dataset →Assay & screening data
High-throughput screens and assay results, including the negatives papers leave out.
Request dataset →Lab notebooks
Medicinal chemistry records showing what was tried, why, and what happened.
Request dataset →Regulatory documents
Study reports and submissions: long, expert-written documents with real stakes.
Request dataset →Safety & pharmacovigilance
Adverse-event reports and safety reviews, fully de-identified.
Request dataset →Hospitals & health
Real care, de-identified.
Clinical data from hospitals and health systems, de-identified at the source and licensed under the hospital's governance.
Health records
Diagnoses, medications, procedures and lab results, linked across each patient's journey.
Request dataset →Clinical notes
Physician notes, discharge summaries and referral letters, with identifiers removed.
Request dataset →Diagnostics & labs
Lab panels and diagnostic test results, with reference ranges and outcomes.
Request dataset →What you get
Licensed once. Usable with confidence.
Clear rights
Chain of title and licence terms documented for every dataset.
Full provenance
Where each file came from, who owns it and how it was prepared.
Built to your spec
Filtered, labelled and formatted for your pipeline, not a raw dump.
Clean evaluation sets
Non-public data means benchmarks your model hasn't already seen.
A data card with every delivery
What is in the set, where it came from, how it was cleaned and what was removed.
Samples before you commit
Review a de-identified sample under NDA before signing a licence.
Tell us what your model needs.
Share the modality, volume and use. If we can source it, we'll come back with a sample. If we can't, we'll say so.