For organisations sitting on data they cannot send to a third-party AI service — proprietary records, patient data, confidential or licensed material, decades of operational history — who want to know what AI can actually do with it, and want a straight answer either way.
Deliverable: a written assessment and a costed recommendation.
What is actually possible with your data, what it would cost to build, and whether it is worth doing. Covers a data inventory and quality assessment, feasibility against your specific problem, architecture options including on-premise requirements, the regulatory and data-governance constraints that apply, and a build-versus-buy recommendation.
We say no when the answer is no. A scoping engagement that ends in "don't build this" has done its job.
Deliverable: a working system on your data, measured results, and an honest verdict.
A focused capability built and evaluated on your own material, so the decision to invest further rests on evidence rather than a demonstration. Includes a requirements workshop, a working prototype running in your environment, an evaluation methodology and measured baseline — what works, what does not, and where it fails — plus integration guidance and a scaling path. Structured metadata and model artefacts remain yours afterwards.
Deliverable: a risk assessment and compliance roadmap, or a governance review report with documentation templates.
We review AI systems against the frameworks that will actually be applied to them — the EU AI Act, GDPR and KVKK, and medical device regulation where relevant — as regulatory-strategy input rather than notified-body work — and identify gaps before they become blockers. Risk classification, data-governance and cross-border transfer review, bias and interpretability assessment, and a prioritised remediation plan.
Deliverable: advisory hours and quarterly reviews.
Fractional AI leadership for organisations building internal capability: monthly advisory hours (8–16 hours), regulatory monitoring, technical architecture review, vendor and model evaluation, and team training.
An engagement begins with scoping — the cheapest way to find out whether the rest is worth doing. Two to four weeks, a written assessment, and a number — including the recommendation not to build, if that is where the evidence lands.
Engagements are founder-led and taken one at a time.
Email contact@gurma.ai with a few lines about the data you hold and the problem you want solved. If it is not something we should take on, we will say so and, where we can, point you somewhere better.