Legal notice
Legal notice and privacy.
Publisher
Coach Drépano, a programme of the Swiss Sickle Cell Association (Route du Levant 7, 1700 Fribourg). Publication director, Prof. David-Z. Issom, Geneva University Hospitals (HUG), david.issom@hug.ch. Prof. Issom leads the Convivens Lab at HEG-Genève (HES-SO), which designs digital infrastructures for equitable, participatory and resilient health systems, with a dedicated focus on rare and neglected conditions.
Hosting
Website and Trévor assistant hosted by Infomaniak Network SA (Geneva, Switzerland), Geneva data centre, ISO 27001 certified. Database never exposed to the Internet: names, messages, answers, notes and watch measurements are encrypted in it, and nightly backups are encrypted and kept for thirty days.
The assistant's answers are produced by a language model operated by Anthropic in the United States, which receives the person's free-text messages, a summary of their follow-up without their first name and any photos or documents they send, after they accept the terms of use or the test agreement. Other providers are involved only if the feature is used: Google Cloud (voice, European endpoints) and Google Wallet (emergency card), OpenStreetMap servers (nearby places), Open-Meteo (weather) and Stripe (online payment).
Personal data
This site works without cookies or trackers. Fonts come from Google Fonts and Instagram thumbnails from Meta's servers, which receive your IP address when loading. Trévor processes your health data with your explicit consent, in accordance with the Swiss Federal Act on Data Protection (FADP) and the GDPR, with the right to access, rectify, export and delete from the “My data” menu or by email.
This data is pseudonymised, not anonymous: Trévor records the first name given, the messaging profile name, the messaging ID (the phone number with WhatsApp) and the answers, but no postal address.
Content and liability
Content is intended for information and education, and cites its sources. Care decisions remain with your medical team.
Trévor complements medical follow-up. It stems from a research project whose evaluation study at HUG is awaiting the ethics committee's authorisation, and remains in development. Trévor is not CE nor FDA cleared or approved. The planned initial consumer experience is for general wellness and is not intended to diagnose, screen for, monitor, treat, cure or prevent disease. Future medical functions require applicable regulatory authorization.
Intellectual property
The Trévor software, its content and the code of this site are the property of David-Z. Issom (Coach Drépano), all rights reserved: the software is proprietary, except for a trust and interoperability core that will be published as open source under the AGPL-3.0 licence (publication in preparation). No licence is granted merely by having access; any reproduction, modification or making available to third parties requires the holder's prior written authorisation, and read access may be granted in writing for an audit. Fonts, logos, clinical protocols, publications and validated questionnaires remain under the terms of their respective holders. Trévor stems from research conducted with Geneva University Hospitals, HES-SO Geneva and the University of Geneva; its first version falls under the invention disclosure filed with Unitec.
Applicable law
Swiss law, place of jurisdiction Geneva. Data supervisory authority, the Federal Data Protection and Information Commissioner (FDPIC).
swiss made software +AI label
swiss made software denotes high-quality Swiss ICT products and values such as quality, innovation and precision in software development. Coach Drépano carries the label with the +AI mention.
AI in Coach Drépano
- Hosting
Trévor's server and its (encrypted) database are hosted in Geneva, at Infomaniak. The language model is operated by Anthropic outside Switzerland (United States): it receives the message, the recent history and a context without first name or identifier, never the full record. - Level
We build on foundation models and add our own layer: an assistant with strict tools, clinical knowledge drawn from HUG protocols, and deterministic safety rules (emergency detection, calculations done by the server). - AI technologies used
Large language models (LLMs) for conversation, reading photos and documents that the person confirms, and translation; a small statistical crisis-risk model, trained on the person's or a cohort's data; a haemoglobin estimation model (HemoSnap) based on a photo of the fingertip, calibrated on the person's own laboratory values, without ever keeping the photo; both models run on our server; Google speech recognition for voice messages. - Model origin and type
Commercial closed-source models from Anthropic (Claude family), used through their API. The crisis-risk and haemoglobin estimation models are our own, centred on the person. We do not operate a proprietary foundation model. - Customisation and fine-tuning
No training or fine-tuning of any language model. Behaviour comes from instructions, knowledge supplied on demand and structured outputs. Under the terms of Anthropic's API, the data sent is not used to train its models. The AI decides nothing alone: the person reviews and confirms before anything is saved.