Data Readiness
How the Hub assesses whether the data behind a use case is available, curated, legally usable, and pipeline-ready – often the true pacing constraint rather than compute.
For many national use cases, the binding constraint is not GPU capacity – it is data. A node can train a model faster than a team can assemble the curated, labelled, legally-usable Nigerian dataset that model needs. The Hub's data-readiness capability makes this constraint visible early, so a use case is not admitted to expensive compute only to stall for want of data.
Why data readiness leads
Compute is now the abundant resource at NAISH; representative local data is the scarce one. Two identified use cases make the point explicitly:
- AI POCUS – the primary constraint is the availability of curated Nigerian ultrasound datasets, not compute. The Hub can train the interpretation model, but only once labelled scans exist.
- Health Record Digitisation – dominant costs are data logistics and storage, not GPU time. The full national backlog could reach hundreds of thousands of GPU-hours and would exceed a single node without phased facility onboarding – a data-sequencing problem as much as a compute one.
The corollary is encouraging: where investment in data exists, readiness is far higher. The AfricanVoices platform – 1,900 hours of curated speech across Hausa, Igbo, Nigerian Pidgin, and Yoruba – means the voice and local-language use cases start with a real corpus rather than a blank slate.
What the assessment covers
- Availability – does the dataset exist at all, and who holds it? Is access negotiated?
- Quality & labelling – is it curated, labelled, and consistent enough to train on, or does it need substantial preprocessing first?
- Legal & consent – is it usable under the Nigeria Data Protection Act (NDPA) and the relevant consent terms, especially for sensitive health and financial records?
- Pipeline maturity – can it be staged into the node's 7.6 TB scratch, processed, and archived within a training window, or does moving it dominate the timeline?
- Representativeness – does it reflect Nigerian populations, languages, dialects, and conditions, so the resulting model serves the whole target group rather than a well-sampled subset?
Where it connects
Data readiness feeds directly into feasibility assessment (a use case with no data path is not feasible yet, regardless of model fit) and into Governance & Risk, where data governance – suitability, legal basis, and retention – sits alongside workload management.
Use-Case Feasibility Assessment
How the Hub evaluates whether a proposed AI use case is technically viable on the node – model fit, GPU intensity, VRAM footprint, concurrency, and the true binding constraint.
GPU / Compute Support
The Hub's core compute capability – shared, governed GPU access on an 8× H200 node, with scheduling, time-slicing, an open-source serving stack, and in-country data residency.