Ideal Customer Profile Image
Product Logo

Maternify — "Chatgpt gives information.Maternify gives her a voice"

A multilingual communication tool that helps minority-ethnic pregnant women in the UK explain concerns, understand NHS communication, and prepare for appointments without diagnosing, replacing clinicians, or pretending to be an interpreter.

Inspiration

The UK has a maternal-health problem hiding in plain sight.

MBRRACE-UK found that 96% of reviewed maternal-death cases had a documented need for an interpreter, yet only 27% actually received one.

We built Maternify around one woman: Maya, a Mandarin-speaking mother navigating pregnancy in the UK. She can hold everyday conversations in English, but during a stressful appointment she struggles to describe symptoms accurately, understand medical terminology, or confidently ask questions.

She receives NHS letters she cannot fully understand. She worries something may be wrong but does not know how to explain it. Most importantly, she lacks the confidence that she is being fully understood.

We realised that the problem was not a lack of information. The problem was communication.

What it does

Maternify focuses on helping women express concerns clearly and understand healthcare communication.

1. Express

Maya describes how she feels using text or voice in Mandarin or English.

The platform generates:

A plain-language explanation in the user's preferred language
A clinically-grounded English summary
A script that can be read or played aloud to a midwife using text-to-speech

This allows women to communicate concerns clearly, even when they struggle to find the right words in English.

2. Symptom Journal

Women can keep a daily record of:

Sleep
Mood
Pain location and severity
Fetal movement
Bleeding
Swelling
Personal notes

Entries are displayed in a horizontally scrollable calendar timeline. Selecting a date expands the input section below the calendar, making it easy to review and update daily observations.

This helps women remember symptoms between appointments and communicate concerns more accurately.

3. Pattern Summaries

Maternify does not diagnose or assess risk.

Instead, it identifies communication-relevant patterns in the information the woman has already recorded.

For example:

"Pain level above 6 reported on five of the last seven days."

When meaningful trends appear, Maternify surfaces communication-focused alerts and generates appointment-ready discussion prompts such as:

"I would like to discuss this with my midwife and understand whether any further assessment is needed."

These prompts can also be played aloud through text-to-speech.

4. Interpret

Users can upload or photograph NHS letters, appointment information, or healthcare documents.

Maternify explains:

What the document means
Any actions required
Important dates
Questions the user may wish to ask at their next appointment

All explanations are grounded in NHS and Tommy's guidance through retrieval-augmented generation (RAG), not open-ended AI responses.

5. Cultural Context Resolution

Many women describe symptoms using familiar cultural expressions, metaphors, or non-clinical language that may not translate clearly into a healthcare setting.

Before generating an English script, Maternify identifies culturally-specific phrases and resolves them into clinically meaningful terminology while preserving the user's original meaning.

For example:

"My stomach feels tight in waves"

becomes

"Possible uterine contractions"

Maternify also performs communication-style calibration, helping users express concerns in a way that is clear, respectful, and appropriate for NHS clinical settings.

This bridges the gap between how symptoms are naturally described within a person's language or culture and how healthcare professionals document and understand those concerns.

The feature currently demonstrates Mandarin as the primary language example, with additional language support planned in future versions.

Safety First

Before any AI processing occurs, a hardcoded safety gate checks for pregnancy red flags such as:

Reduced fetal movement
Heavy bleeding
Severe headache with visual disturbance

If detected, the system immediately displays emergency guidance and contact options for maternity triage or emergency services.

The AI is never allowed to override or dismiss a red-flag situation.

How We Built It

Next.js 14 with App Router
TypeScript
Tailwind CSS
Vercel deployment
GLM-5.1 via Z.ai for language generation
Google Gemini for speech-to-text, text-to-speech, OCR, and cultural-context processing
Supabase for storing de-identified usage data
RAG system built on curated NHS and Tommy's pregnancy guidance
Zod validation for every AI response before it reaches the user

The application is presented as a mobile-first experience within a browser-based phone simulator to demonstrate how it would function as a future native mobile app.

Challenges We Faced

The biggest challenge was staying on the safe side of the line between a communication tool and a medical device.

We wanted to help women describe symptoms more effectively without creating a symptom checker or diagnosis engine.

Other challenges included:

Handling Mandarin language variations and culturally specific symptom descriptions
Translating idiomatic expressions into clinically meaningful terminology without changing the user's intent
Detecting pregnancy red flags consistently across multiple languages
Ensuring every AI response remained grounded in NHS-approved information
Creating reliable speech, OCR, and audio playback workflows suitable for a live demo environment

To address these challenges, we implemented deterministic red-flag detection, strict schema validation, source-grounded retrieval, and fallback mechanisms for every AI-generated response.

What We Learned

The most impactful healthcare AI is not necessarily the most powerful model.

It is the model with the clearest boundaries.

Women often do not need another tool telling them what condition they might have. They need help expressing themselves clearly, understanding what healthcare professionals are telling them, and feeling confident enough to participate in their own care.

Building constraints into the system ultimately made the product safer, more trustworthy, and more useful.

We also learned that language barriers are not simply translation problems. Cultural expressions, communication styles, and confidence gaps can be just as significant as vocabulary itself.

What's Next

Our next steps include:

Expanding language support to Urdu and Bengali
Building native iOS and Android applications
Creating clinician-readable communication timelines
Supporting secure consent-based sharing with maternity teams
Exploring NHS App integration
Developing the regulatory and clinical-safety pathway required for real-world deployment

Maternify does not try to replace healthcare professionals.

It simply ensures that every mother has the ability to be heard, understood, and involved in her own care.