Perinatal mental health support that keeps sensitive data private by design
A support app for mothers and fathers that puts our doctoral research on privacy-preserving conversational AI into practice.
- Owner
- DesMind Labs
- Sector
- Digital health
- Services
- Research, AI and privacy engineering, full-stack
- Year
- 2026
What needed solving
Perinatal mental health difficulties affect both parents, yet many signs go unnoticed until they become serious. Digital support can reach more families, but conversations about mental health are among the most sensitive data a person can share.
Our scoping review of AI conversational agents in this field found that privacy, ethics and inclusivity are often treated as afterthoughts. PeriParen was designed the other way round.
The solution
Support for both parents
A dyadic design: parents invite their partner, so fathers and partners are included, not only mothers.
Validated screening
Standard questionnaires such as the EPDS and PHQ, alongside regular wellbeing check-ins.
Controlled disclosure
Each parent decides which insights their partner can see. Nothing is shared by default.
Conversational support
A chat assistant for guidance between check-ins. Wellbeing insights and audit metadata are kept, not raw conversations.
Separated data stores
Identity and wellbeing data live in two separate databases.
Replaceable language model
The LLM sits behind a provider interface, so models can be changed without touching the app.
How the system fits together
A simplified view of the main components and how requests move between them.
- React web appCheck-ins, conversations, insights
- FastAPI serviceAuth, screening, check-ins, partner linking, chatLLM provider interfaceSwappable model back end
- Identity databaseAccounts and partner links, PostgreSQLWellbeing databaseDerived insights and audit metadata, PostgreSQL
Choices that made the difference
Identity and wellbeing data are kept apart
Two logical databases mean that a breach of one does not link personal details to mental health information.
No raw transcripts are stored
Only audit metadata and derived insights are kept, which minimises what could ever be exposed.
The model is replaceable
A provider interface keeps the app independent of any single AI vendor, which matters for privacy reviews and cost.
Research first, product second
The design follows the evidence and gaps identified in our peer-reviewed scoping review.
PeriParen is a working MVP within DesMind Labs. We are looking for clinical and research partners to run a pilot.
Technology
Have a system like this in mind?
Tell us what you are trying to achieve. We reply within two business days.