AI transparency notice

What H2Hub's AI features do, which models power them, what data they receive, where a human stays in control and where they are known to be weak.

Last updated 9 August 2026

Document control

Version: 1.1
Last reviewed: 9 August 2026
Next scheduled review: 9 February 2027
Scope: AI system inventory

Reviewed on change as well as on schedule: this document is re-issued whenever an AI model, AI feature, subprocessor, hosting region or transfer mechanism changes.

Our role

H2Hub is a deployer of third-party general-purpose AI models, accessed through an API gateway. We do not train, fine-tune or develop foundation models. We design prompts, supply context and present results inside our product.

You always know when you are reading AI-generated content: it is labelled in the interface and shown next to the underlying figures or datasets so you can check the basis for it.

Intended purpose and prohibited uses

Intended purpose: assisting technical and commercial teams with preliminary hydrogen project-development analysis, explanation and education. Outputs are decision-support material for professional users.

H2Hub's AI features are not designed or permitted to be used for:

assessing, scoring, profiling or making decisions about individual people;
employment, credit, insurance, education-admission or law-enforcement decisions;
producing construction, permitting or regulatory-submission documentation;
final investment decisions without independent professional review;
safety-critical control of physical equipment.

Classification and review

On our own assessment, H2Hub's AI features are not prohibited practices and do not fall within the EU AI Act's high-risk annexes: they analyse hydrogen projects (capacity, cost, resource and schedule assumptions) rather than making decisions about people, employment, creditworthiness, education, essential services or law enforcement. We treat our obligations as those of a deployer of general-purpose AI, plus transparency: AI content is labelled, its purpose and limits are documented here, and a human decides what to do with it.

This inventory is reviewed whenever a feature, model or provider changes, and at least annually. Records of the assessment are kept by the controller and made available to enterprise customers on request.

Model-calculated output vs. AI-generated explanation

H2Hub keeps two things strictly separate, in the interface and in exports:

Model-calculated output — produced by H2Hub's deterministic engines (site screening, physical yield and dispatch, LCOH and economics, stress tests, power-pathway comparison). The same inputs always produce the same numbers. Assumptions, units, data provenance and the calculation version are recorded, and results are reproducible. No language model produces or alters these figures.
AI-generated explanation — narrative interpretation of numbers that already exist: summaries, rationales, teaching text, suggested next steps. It is labelled as AI-generated, shown next to the figures it describes, may contain errors, and never changes a calculated result.

Neither is a recommendation. H2Hub reports what its models compute; it does not tell you to invest in, build, permit or finance anything, and its outputs are preliminary decision-support material rather than licensed engineering, financial or investment advice.

Human oversight

Every AI output is advisory and reversible: nothing is committed, submitted or executed automatically. The user reviews, edits and decides. Exports carry a preliminary-analysis stamp, and no AI feature applies a score or narrative to a person.

AI system inventory

H2Hub AI project assistant (chat)

In-app assistant and course Q&A

Intended purpose: Explain hydrogen engineering and project-development concepts and help users navigate the 7-stage path.
Model / provider: Lovable AI Gateway → OpenAI GPT class model
Inputs (data sent to the model): The user's typed question and the page context. No account credentials or payment data.
Outputs: Free-text explanations and navigation suggestions. No score, decision or automated action.
Affected users: The signed-in professional user who asked the question. No third parties are assessed.
Personal data handled: Non-personal technical context by design. Personal data only if the user types it into the prompt; users are asked not to.
Human oversight: Answers are informational only; the user decides whether to act. Every answer sits next to the preliminary-analysis notice.
Known limitations: May make factual or analytical errors, may be out of date, may misread ambiguous questions, and does not verify site-specific conditions. Outputs require verification.

Project advancement explanation

Stage 2 Screen / Stage 7 Prepare narratives

Intended purpose: Turn model outputs and dataset comparables into a plain-language rationale for a project's development status.
Model / provider: Lovable AI Gateway → Google Gemini class model
Inputs (data sent to the model): Non-personal project attributes: capacity, technology, location, status, dataset comparables.
Outputs: A short written rationale shown beside the underlying figures.
Affected users: The project owner and invited collaborators. The subject of the analysis is a project, not a person.
Personal data handled: Project attributes only. No personal data is sent.
Human oversight: Rendered as a rationale beside the underlying numbers so the user can check the basis.
Known limitations: Narrative quality depends on comparables available for the region; may over-generalise where peer data is thin. The user remains responsible for assumptions and decisions.

Predictive analytics / success indicators

Intelligence and readiness panels

Intended purpose: Score project records against historical outcome patterns to indicate relative development momentum.
Model / provider: In-house statistical model plus AI-generated summaries via Lovable AI Gateway
Inputs (data sent to the model): Aggregated project-level dataset attributes. No personal data.
Outputs: An indicative band or score for a project record, plus a written summary of the drivers.
Affected users: Users reading the panel. Scores describe public project records, never individuals.
Personal data handled: Public and aggregated project data. No personal data.
Human oversight: Scores are labelled as indicative, always shown with the band definitions, and never presented as a decision.
Known limitations: Trained on historical public project records; regions or technologies with few records produce weak signals. Not a prediction of commercial outcome.

Research and market summaries

Intelligence research generation

Intended purpose: Summarise public market and dataset information into briefing text.
Model / provider: Lovable AI Gateway → OpenAI GPT class model
Inputs (data sent to the model): Public dataset extracts and the user's research prompt.
Outputs: Draft briefing text with citations to the datasets used.
Affected users: The requesting user; drafts may be shared internally by that user.
Personal data handled: Public dataset extracts and the prompt text. No personal data by design.
Human oversight: Output is a draft for the user to verify against the cited datasets.
Known limitations: Can omit or compress nuance; citations must be checked before reuse.

Data used by AI features

Prompts and context are sent to the model providers listed above solely to generate your result. Under our API-tier agreements, prompts and outputs are not used to train the providers' models. Processing takes place in the United States under Standard Contractual Clauses. Prompt and response logs are retained for up to 30 days for abuse prevention and debugging, then deleted. See the EU & UK compliance notice and the project data confidentiality policy.

Reporting a problem

If an AI output looks wrong, unsafe or inappropriate, email contact@reneenergy.com with the feature name and, where possible, the text you saw. We review reports, correct prompts or context, and update this notice when a feature or model changes.