How it works
Three ways we work with corporate corrosion teams. Bespoke engagements take two shapes: we apply AI to a defined problem on your data and hand over an auditable result, or we build a custom AI tool — shaped around your data and workflows — that your team uses independently going forward. Corporate training extends the same methodology into your team's own capability, covering AI literacy and AI for corrosion R&D in your specific scientific context.
Bespoke engagements
Deliverable is either an auditable result on your data, or a custom tool you keep.
Read the bespoke methodology →Open demos
Working demos of methodology you can break before any engagement. The Experts Discussion runs a multidisciplinary panel on your question; Expert Creation lets you build an expert from your own sources to query or add to a discussion.
- EIS Fitting — Equivalent-circuit fits with parameter ranges and residuals — one spectrum or a full archive.
- Experts Discussion — A multidisciplinary roundtable, grounded in our knowledge or in yours.
- Expert Creation — Build an expert from your own sources — yours to query or to add to a discussion.
Corporate training
Hands-on AI literacy for R&D teams, scoped to your specific scientific context — not generic.
Looking for public catalogue courses instead? See /courses.
Read about corporate training →Methodology principles
- 01
AI as engine, not solver.
We don't hand a problem to a model and accept what comes out. We decompose each problem into a knowledge-driven workflow with well-defined steps, then use the model inside specific steps where it fits — with domain knowledge injected to constrain what it produces.
- 02
Each step is named and documented.
Every step in the workflow names the method it uses and the assumptions it starts from — algorithm classes, model families, retrieval schemes, priors — written down where a reviewer can read them. Where a closed-source component sits in the path, we say so.
- 03
Each step produces an inspectable artifact.
The workflow surfaces an intermediate output at every step — a proposal, an annotation, a draft, a fit. You don't see only the final answer; you see how the answer was assembled, in the order it was assembled.
- 04
Uncertainty travels with the artifact.
Each output carries the information needed to judge it — error ranges, confidence signals, source references — attached at the step that produced them. A point estimate without context is not a deliverable.
- 05
Humans review and can override at every gate.
Engineers can review each intermediate artifact and can accept, reject, edit, or rerun the step. The final result is a chain of human judgements over AI-proposed artifacts, not a single model output we ask you to trust.
- 06
Methods trace to a published record.
Every method we apply traces to a published precedent or our own peer-reviewed work. The lineage is named in product documentation and on /about/science, where the founder's research record lists the papers behind the methods.