maya, multilingual onboarding
Care organizations in Germany are facing a structural crisis: 350,000 nursing staff will be missing by 2034, and a growing share of trainees enter with language gaps that current onboarding cannot close. Each dropout costs organizations an average of €6,826.
maya is an enterprise onboarding system designed with Stiftung Liebenau (8,900+ employees) that addresses language friction at its root — before it becomes a workplace problem. Two linked systems: an adaptive AI assistant that evolves with the trainee, and a structured co-living matching model that shapes language habits before the first shift.
Projected outcomes: €131/month saved per trainee in staff overhead, 3.75 hours of coordination time recovered per trainee per month.
My role
Product design research synthesis interaction and UI prototyping
Scope
Research -> prototype System concept
Domain
Enterprise Onboarding Workflow Systems Language Accessibility
Tools
Figma Claude Code
Team
Luke Caporelli Jan Lonardoni Lukas Predan Danlei Fu Peter Schneider
Year
2025
context
Settings and stakes
Care organizations depend on fast, precise communication across shifts and teams. The partner context required privacy, cultural sensitivity, and adoption from day one.
Core problem
International trainees often have formal language certificates but still struggle with everyday routines, small talk, and real workplace phrasing. This creates repeated explanations, time loss, and avoidable misunderstandings.
Why the usual fixes fail
Courses and one off trainings rarely translate into reliable performance under daily work pressure. Without continuous, in context support, the gap stays.
Partner: stiftung liebenau
8,900+
Employees
Research signals
6,826 €
Direct net cost per apprenticeship dropout
68,000
Apprentices with foreign citizenship in German health and care occupations
350,000
Nursing staff potentially missing in Germany by 2034
Discover
Research approach
We combined desk research, focus group interviews, on site visits, and stakeholder input to capture both lived reality and organizational constraints.
What we heard repeatedly
Trainees avoid asking questions, feel uncertain in routines, and default to their native language in shared living. Staff carry recurring re explanation work and coordination overhead.
Key insight
The language problem starts before the job, inside everyday life. If daily routines stay mono language, the workplace inherits the deficit.
define
Target direction
Build everyday language capability through low-pressure daily life — not exam-driven enforcement.
Problem framing
Language is the upstream lever that amplifies satellite issues such as onboarding overload, missing local knowledge, housing friction, and staff burden. We focused on language because it unlocks multiple downstream wins.
Success criteria
Trust and adoption are the success gates in this domain. The product must feel safe, respectful, and operationally realistic for high workload teams.
Problem framing and direction
Problem map
Language is the upstream lever that drives multiple downstream failures.
Why existing solutions fail
Courses do not transfer into daily care communication. Training needs in context support.
Design principles
Mutual learning, help for self help, everyday independence first.
System rationale
Why shared housing is a leverage point
Daily language exposure happens at home, not in class. Matching shapes habits before the first shift.
Peer learning built into the system
Pair advanced German speakers with beginners to turn everyday routines into low pressure practice.
One assistant, staged maturity
Same assistant evolves from onboarding clarity to everyday support as confidence grows.
ideate
System concept
maya is a living and integration platform combining shared living group matching (shared housing) with digital guidance for onboarding and everyday life, designed to reduce dependency and lower coordination load.
Two moments that matter
Pre arrival and early onboarding, when uncertainty is highest. Then the first months in daily life, when routines form and language habits lock in.
Language learning mechanism
maya starts in the native language and shifts gradually toward German, enabling passive exposure and active practice without forcing artificial learning moments.
Language Integration Levels -> Co-Living Matching Structures
prototype
AI Assistant, Phase 1 to 2
One assistant, two phases. Phase 1 handles onboarding questions to reduce repetitive explanations. Phase 2 shifts into everyday support — routines, practical questions — within a defined, safe-first scope.
Shared living groups matching system
Matching pairs advanced German speakers with beginners to create everyday language exposure through shared living.
Internal Dashboard Concept
A provider facing dashboard concept offers overview and decision support, so recurring questions and support needs can be identified and handled systematically rather than ad hoc.
matching flow
daily helper flow
matching flow
deliver
What we shipped as a semester outcome
A coherent product narrative and three linked prototypes: onboarding assistant, everyday assistant with adaptive language shift, and shared living groups matching logic.
Domain constraints in the first phases
Early phases intentionally exclude medical data and clinical workflows. We focused on onboarding and daily life support to build trust and adoption first.
Operational readiness signals
Privacy, cultural tone, and acceptance shaped the interaction model, escalation logic, and scope boundaries as primary design constraints.
prototype
impact
Expected user impact
Less uncertainty in daily routines, fewer avoidance moments, and faster confidence. The assistant lowers the social cost of asking questions and supports steady learning.
Expected organizational impact
Reduced staff overhead from repeated explanations, fewer coordination loops, and better housing utilization through structured matching.
Economic framing
We used a simplified model to communicate the order of magnitude: support time, fully loaded staff cost, and dropout cost reference values, framed as expected impact rather than measured outcomes.
Research signals
€131 / month
Expected savings per trainee
3.75 h / month
Expected STAFF TIME SAVED PER TRAINEE
10%
Expected decrease in dropouts (assumption)
What changed my thinking
Trust constraints
In care, trust is the feature. Privacy and tone decide adoption.
Next build targets
Separate private and work assistants. Add source clarity and escalation paths.
Principle statement
Prevent language breakdown early, or pay for rework later across the whole system.
reflection
What I learned
Enterprise problems rarely have interface solutions. The real design work was understanding organizational constraints — trust, adoption thresholds, and operational feasibility — before touching any screen. In high-stakes care environments, a product that doesn't get adopted is the same as no product.
What I would improve next
Make escalation paths and source transparency more explicit in the UI, and separate private life and work context assistants to reduce ambiguity for users.
What I would build if expanded
After trust is established, deeper workflow integration as a later phase — with strict governance and access control, aligned to a phased adoption roadmap.
maya, multilingual onboarding
maya reduces everyday language friction in care environments through two linked systems: an onboarding assistant plus a gradual, adaptive language shift from native language to German, supported by structured shared living matching.
context
Settings and stakes
Care organizations depend on fast, precise communication across shifts and teams. The partner context required privacy, cultural sensitivity, and adoption from day one.
Core problem
International trainees often have formal language certificates but still struggle with everyday routines, small talk, and real workplace phrasing. This creates repeated explanations, time loss, and avoidable misunderstandings.
Why the usual fixes fail
Courses and one off trainings rarely translate into reliable performance under daily work pressure. Without continuous, in context support, the gap stays.
Partner: stiftung liebenau
8,900+
Employees
Discover
Research approach
We combined desk research, focus group interviews, on site visits, and stakeholder input to capture both lived reality and organizational constraints.
What we heard repeatedly
Trainees avoid asking questions, feel uncertain in routines, and default to their native language in shared living. Staff carry recurring re explanation work and coordination overhead.
Key insight
The language problem starts before the job, inside everyday life. If daily routines stay mono language, the workplace inherits the deficit.
Research signals
6,826 €
Direct net cost per apprenticeship dropout
68,000
Apprentices with foreign citizenship in German health and care occupations
350,000
Nursing staff potentially missing in Germany by 2034
define
Target direction
Build everyday language capability through low-pressure daily life — not exam-driven enforcement.
Problem framing
Language is the upstream lever that amplifies satellite issues such as onboarding overload, missing local knowledge, housing friction, and staff burden. We focused on language because it unlocks multiple downstream wins.
Success criteria
Trust and adoption are the success gates in this domain. The product must feel safe, respectful, and operationally realistic for high workload teams.
Problem framing and direction
Problem map
Language is the upstream lever that drives multiple downstream failures.
Why existing solutions fail
Courses do not transfer into daily care communication. Training needs in context support.
Design principles
Mutual learning, help for self help, everyday independence first.
ideate
System concept
maya is a living and integration platform combining shared living group matching (shared housing) with digital guidance for onboarding and everyday life, designed to reduce dependency and lower coordination load.
Two moments that matter
Pre arrival and early onboarding, when uncertainty is highest. Then the first months in daily life, when routines form and language habits lock in.
Language learning mechanism
maya starts in the native language and shifts gradually toward German, enabling passive exposure and active practice without forcing artificial learning moments.
System rationale
Why shared housing is a leverage point
Daily language exposure happens at home, not in class. Matching shapes habits before the first shift.
Peer learning built into the system
Pair advanced German speakers with beginners to turn everyday routines into low pressure practice.
One assistant, staged maturity
Same assistant evolves from onboarding clarity to everyday support as confidence grows.
Language Integration Levels ->
Co-Living Matching Structures
prototype
AI Assistant, Phase 1 to 2
One assistant, two phases. Phase 1 handles onboarding questions to reduce repetitive explanations. Phase 2 shifts into everyday support — routines, practical questions — within a defined, safe-first scope.
Shared living groups matching system
Matching pairs advanced German speakers with beginners to create everyday language exposure through shared living.
Internal Dashboard Concept
A provider facing dashboard concept offers overview and decision support, so recurring questions and support needs can be identified and handled systematically rather than ad hoc.
daily helper flow
deliver
What we shipped as a semester outcome
A coherent product narrative and three linked prototypes: onboarding assistant, everyday assistant with adaptive language shift, and shared living groups matching logic.
Domain constraints in the first phases
Early phases intentionally exclude medical data and clinical workflows. We focused on onboarding and daily life support to build trust and adoption first.
Operational readiness signals
Privacy, cultural tone, and acceptance shaped the interaction model, escalation logic, and scope boundaries as primary design constraints.
matching flow
prototype
impact
Expected user impact
Less uncertainty in daily routines, fewer avoidance moments, and faster confidence. The assistant lowers the social cost of asking questions and supports steady learning.
Expected organizational impact
Reduced staff overhead from repeated explanations, fewer coordination loops, and better housing utilization through structured matching.
Economic framing
We used a simplified model to communicate the order of magnitude: support time, fully loaded staff cost, and dropout cost reference values, framed as expected impact rather than measured outcomes.
Research signals
€131 / month
Expected savings per trainee
3.75 h / month
Expected STAFF TIME SAVED PER TRAINEE
10%
Expected decrease in dropouts (assumption)
reflection
What I learned
Personalization only works when routing decisions are transparent and explainable.
What I would improve next
Make escalation paths and source transparency more explicit in the UI, and separate private life and work context assistants to reduce ambiguity.
What I would build if expanded
After trust is established, explore deeper workflow integration as a later phase with strict governance and access control, aligned with a phased roadmap.
What changed my thinking
Trust constraints
In care, trust is the feature. Privacy and tone decide adoption.
Next build targets
Separate private and work assistants. Add source clarity and escalation paths.
Principle statement
Prevent language breakdown early, or pay for rework later across the whole system.