Aidenn Model Card (v1.5)

Aidenn v1.5 (Property-Repair Diagnostic and Triage Model)

Aidenn is a multimodal AI system used within the Help me Fix platform to analyse user-reported symptoms (text, images, and video) and suggest possible causes for property-repair issues such as heating faults or leaks. The model may use large language model components to generate questions and summarise issues. Aidenn does not make binding decisions about whether repairs will be carried out; a qualified engineer always reviews its output.

Aidenn is intended to assist tenants and property-managers in diagnosing property-repair issues quickly and preparing for a first-time fix. Outputs should always be reviewed by a qualified engineer before any physical work is undertaken.

Residents of managed properties, letting agents, housing providers, and authorised engineers aged 18+. The model is not intended for minors or for diagnosing health-related issues.

Aidenn is trained on a curated dataset of anonymised property-repair reports, manufacturer manuals, publicly available troubleshooting guides, and synthetic data generated by subject-matter experts. Video clips used for AI training are anonymised and stripped of personally identifiable information. Data from EU/UK residents are only used for training with consent or on the basis of a recognised legitimate interest (e.g., improving essential repair services). Sensitive data (e.g., biometric or emotional data) are excluded.

Personal data is not used to improve Aidenn; anonymised repairs data is stored separately from identifiable data. Retention periods are limited to what is necessary (e.g., two years), and deletion or anonymisation is performed promptly upon request.

Aidenn is evaluated and scored on diagnostic accuracy (precision/recall against ground truth fault categories) by a human-in-the-loop system which is facilitated by a qualified trade professional.

Aidenn must not discriminate against tenants; it does not infer protected characteristics or make decisions on tenancy continuation or rent. Human engineers supervise all outcomes, and users can bypass the AI process by reporting their issue via telephone or email to their property manager. The model avoids storing unnecessary personal information and does not perform emotion recognition or biometric categorisation.

The model version and performance metrics are documented and updated at least annually or after major re-training. Historic models are retained for audit.

Users can provide feedback via the Help me Fix support portal or by email to the Data Protection Officer (DPO) at legal@tradeinnovations.group. Complaints about AI decisions will trigger human review.