Trust architecture for modern home care

Caire automates 50%+ of manual scheduling work, pushes staff efficiency to 75–80%, and protects continuity for tens of thousands of clients. Explore how the platform runs end-to-end—from data ingestion and optimization to enterprise-grade security.

Proven production metrics from Swedish municipal deployments show how the Caire platform transforms operational, financial, and human outcomes simultaneously.

Each optimization run produces better schedules, happier staff, and more predictable outcomes. Data from every execution feeds the intelligence layer, strengthening the platform for all providers on the network.

• Executives: Realized 75–80%+ care time and transparent ROI dashboards.
• Schedulers: AI handles constraints; planners focus on strategic choices.
• Caregivers: Shorter routes, protected breaks, and continuity for clients.
• Municipalities: Lower emissions, better use of public budgets, higher trust.

The platform eliminates spreadsheet firefighting and unlocks strategic execution.

The evolution of home care operations from Level 0 to Level 4.

• Level 0: Phone & Spreadsheets (Manual)
• Level 1: Digital Planning – Static plans, manual adjustments, no optimization.
• Level 2: Assisted Scheduling (Live today) – Optimization engine, daily adjustments. Advanced optimization (manual trigger). Produces ready-to-publish proposals.
• Level 3: Self-Driving Scheduling (Roadmap) – Autonomous handling of disruptions. Real-time optimization (streaming GPS, cancellations, traffic). Planners only review exceptions.
• Level 4: Predictive Operations (Roadmap) – Autonomous operations, forecasting using data and ML pipelines.

A unified architecture supporting all levels of automation—from assisted planning to predictive operations.

flowchart LR %% ========= COLORS (Caire palette-inspired placeholders) ========= classDef source fill:#cce7ff,stroke:#2a7de1,color:#000; classDef storage fill:#e0ffe5,stroke:#31a353,color:#000; classDef process fill:#fff3c4,stroke:#c79b00,color:#000; classDef security fill:#ffe0e0,stroke:#d63a3a,color:#000; classDef output fill:#e4d7ff,stroke:#7a4ed8,color:#000; %% ========= DIAGRAM START ========= subgraph "Attendo Systems (eCare Welfare)" welfareAPI["Welfare API (REST / Authentication)"]:::source welfareCSV["Welfare CSV Export (optional)"]:::source welfarePDF["PDF: Care Decisions (manual upload)"]:::source end subgraph "Caire Platform (AWS Stockholm - EU/EEA)" ingest["Data Ingestion Layer (API Gateway / Import Services)"]:::process validate["Validation + Normalization (Business Rules)"]:::process encrypt["Encryption & Access Controls (Transport & At-Rest)"]:::security store["Datastore

Planning Data (Visits, Time Windows, Addresses, Skills/Tags) Staff Data (Shifts, Availability, Skills, Employment Type, Geography) Case/Welfare Data (Needs, Care Plans, Restrictions) Location Data (Home addresses, Units, Zones)

• Planning Data (Visits, Time Windows, Addresses, Skills/Tags)
• Staff Data (Shifts, Availability, Skills, Employment Type, Geography)
• Case/Welfare Data (Needs, Care Plans, Restrictions)
• Location Data (Home addresses, Units, Zones)

API ingest (preferred) CSV import (secondary) Secure PDF uploads (care plans) Validation + Schema Mapping

• API ingest (preferred)
• CSV import (secondary)
• Secure PDF uploads (care plans)
• Validation + Schema Mapping

Phase 1: Demand Aggregation – Merge visits by time/route/geo, Identify batchable windows Phase 2: Candidate Assignment Generation – Match employees → visits by constraints, Filter by role, skills, continuity, travel feasibility Phase 3: Optimization – Routing solver (minimize travel time), Compression objective (reduce idle time), Priority: continuity → employee preference → cost → travel Scheduler Experience

• Phase 1: Demand Aggregation – Merge visits by time/route/geo, Identify batchable windows
• Phase 2: Candidate Assignment Generation – Match employees → visits by constraints, Filter by role, skills, continuity, travel feasibility
• Phase 3: Optimization – Routing solver (minimize travel time), Compression objective (reduce idle time), Priority: continuity → employee preference → cost → travel

Caire auto-resolves sick calls, route conflicts, and gaps in minutes. Planners start each morning with validated proposals and can launch what-if scenarios with one click.

• Morning startup: 15–20 minutes to review AI proposals.
• Intraday: real-time optimization flips reactive phone trees into proactive alerts.
• Monthly: ghost “ideal” schedules highlight when it’s time to re-baseline the roster.

The recommendation engine guides planners from the first municipal decision to continuous optimization:

• Onboard: AI reads care plans and suggests ideal visit patterns and competencies.
• Approve: Clients and planners adjust preferences; the solver rebalances instantly.
• Operate: Daily automation keeps schedules feasible as reality shifts.
• Improve: Monthly baseline comparisons expose when efficiency drifts, prompting re-optimization.

Conservative continuity, balanced operations, aggressive savings—the continuous learning loop runs them all on the same data so leaders can select the path that matches policy and budgets.

Protects continuity at all costs. Useful during change management and sensitive client cohorts.

Recommended daily driver weighting travel, efficiency, and continuity for steady 75–80% performance.

Pushes costs and wait times down when staffing is tight—often used for special projects.

Fine-tune weights per municipality, area, or contract. Save templates for recurring scenarios.

The platform blends modern web tech, optimization engines, and secure EU infrastructure.

• Frontend: Next.js 15, React 19, Tailwind, shadcn/ui, professional calendar components.
• Backend: Next.js API routes, Drizzle ORM, Redis cache, AWS EC2 (EU).
• Auth & Security: Clerk v6 SSO/MFA, organization-scoped RBAC, HTTPS/HSTS, audit trails.
• DevOps: GitHub Actions CI/CD, PM2 clustering, CloudWatch, Sentry, automated DR.

Caire uses a deterministic "White Box" approach, ensuring full transparency and control.

• No "Black Box": Our optimization engine is rule-based and deterministic. No hallucinations or hidden training biases.
• Human in the Loop: The system is a decision-support tool. Planners always retain final approval.
• Data Privacy: No customer data is used to train external models. Your data stays yours.

A transparent, secure, and compliant foundation for modern home care. We ensure your data remains yours—never used to train public models.

Input: Data is ingested via secure APIs (e.g., Carefox), encrypted CSV imports, or direct input. Import is manual — triggered when creating a schedule, not on an automatic timer. Storage: All data resides in EU-hosted PostgreSQL databases with strict tenant isolation (Row Level Security). Processing: The AI engine optimizes schedules in memory using mathematical rules. No customer data is used for training. ( See Caire Core scheduling and routing technical overview ) Output: Optimized schedules are presented for human review and can be exported back to source systems. EU AI Act

• Input: Data is ingested via secure APIs (e.g., Carefox), encrypted CSV imports, or direct input. Import is manual — triggered when creating a schedule, not on an automatic timer.
• Storage: All data resides in EU-hosted PostgreSQL databases with strict tenant isolation (Row Level Security).
• Processing: The AI engine optimizes schedules in memory using mathematical rules. No customer data is used for training. ( See Caire Core scheduling and routing technical overview )
• Output: Optimized schedules are presented for human review and can be exported back to source systems.

Low Risk Classification: CAIRE is an operational optimization tool, not a "High Risk" system under the EU AI Act. Explainable by Design: No "black box" neural networks. Every decision is traceable to configured rules and weights. Data Sovereignty: You own your data. We do not use it to train generative models or share it with third parties. Human-in-the-Loop: The system is designed to support human planners, not replace their final judgment.

• Low Risk Classification: CAIRE is an operational optimization tool, not a "High Risk" system under the EU AI Act.
• Explainable by Design: No "black box" neural networks. Every decision is traceable to configured rules and weights.
• Data Sovereignty: You own your data. We do not use it to train generative models or share it with third parties.
• Human-in-the-Loop: The system is designed to support human planners, not replace their final judgment.

Designed for Swedish municipalities with EU residency, GDPR controls, and healthcare data governance from day one.

All data hosted in EU via Amazon RDS (Stockholm region). Organization ID enforced on every query and cache lookup. Nightly backups with point-in-time recovery and retention policies. Full audit trails for schedule changes and optimization runs.

• All data hosted in EU via Amazon RDS (Stockholm region).
• Organization ID enforced on every query and cache lookup.
• Nightly backups with point-in-time recovery and retention policies.
• Full audit trails for schedule changes and optimization runs.

Clerk MFA, SSO, and per-role permission enforcement. HttpOnly, SameSite, and Secure cookies; CSRF protection. CloudWatch + Sentry alerts for anomalies and degradation. Pen-test ready: layered defenses and sanitized inputs. Implementation Model

• Clerk MFA, SSO, and per-role permission enforcement.
• HttpOnly, SameSite, and Secure cookies; CSRF protection.
• CloudWatch + Sentry alerts for anomalies and degradation.
• Pen-test ready: layered defenses and sanitized inputs.

Caire hides the complexity but never removes oversight. Teams approve changes, run what-if experiments, and export back to source systems on their terms.

• Nightly automation prepares fresh schedules before the workday starts.
• Schedulers review, adjust, and publish via unified dashboards in minutes.
• Leadership monitors KPIs and ROI in real time to steer investments.
• Level 3 Roadmap: Continuous learning improves route heuristics, continuity weights, and forecasts.

Technical details on how recurring patterns (slingor) and visit pinning enable stable yet flexible scheduling.

slinga: Stores recurring weekly patterns per caregiver, weekday, and shift. Includes version tracking for pattern evolution. slinga_visits: Individual visits within a slinga pattern, including planned times, sequence, and pinning status. visits: Extended with slingaId , pinned (boolean), and assignment fields ( assignedEmployeeId , assignedStartTime , assignedEndTime ). solution_visits: Candidate assignments from optimization jobs, allowing comparison before acceptance. Pinning Mechanics

• slinga: Stores recurring weekly patterns per caregiver, weekday, and shift. Includes version tracking for pattern evolution.
• slinga_visits: Individual visits within a slinga pattern, including planned times, sequence, and pinning status.
• visits: Extended with slingaId , pinned (boolean), and assignment fields ( assignedEmployeeId , assignedStartTime , assignedEndTime ).
• solution_visits: Candidate assignments from optimization jobs, allowing comparison before acceptance.

Pinned visits: Marked with pinned=true , these visits cannot be moved by the optimization engine. They maintain continuity and stability. Unpinned visits: Marked with pinned=false , these visits can be optimized by AI to improve efficiency, travel time, and workload balance. Pinning rules: When a visit is pinned, earlier conflicting visits in the same shift must also be pinned to satisfy optimization constraints. Assignment flow: Visits start unpinned, get optimized, then become pinned when accepted by planners. From-Patch API

• Pinned visits: Marked with pinned=true , these visits cannot be moved by the optimization engine. They maintain continuity and stability.
• Unpinned visits: Marked with pinned=false , these visits can be optimized by AI to improve efficiency, travel time, and workload balance.
• Pinning rules: When a visit is pinned, earlier conflicting visits in the same shift must also be pinned to satisfy optimization constraints.
• Assignment flow: Visits start unpinned, get optimized, then become pinned when accepted by planners.

Full solve endpoint: Used for cold starts and major redesigns. Sends complete schedule with all visits and pinning flags. From-patch endpoint: Used for incremental updates. Sends only changed visits, inherits previous solution and pinned assignments. Use cases: Real-time disruptions (Scenario C), adding new clients (Scenario B), fine-tuning daily schedules. Performance: From-patch is faster because it reuses previous solution as starting point, typically completing in seconds vs minutes for full solve. Data Flow

• Full solve endpoint: Used for cold starts and major redesigns. Sends complete schedule with all visits and pinning flags.
• From-patch endpoint: Used for incremental updates. Sends only changed visits, inherits previous solution and pinned assignments.
• Use cases: Real-time disruptions (Scenario C), adding new clients (Scenario B), fine-tuning daily schedules.
• Performance: From-patch is faster because it reuses previous solution as starting point, typically completing in seconds vs minutes for full solve.

Import: External schedules converted to slingor with visits marked pinned=true . Daily expansion: Published slingor expanded into concrete visits for each date, all marked pinned. Optimization: Full schedule sent to optimization engine. Pinned visits stay fixed, unpinned visits can move. Solution storage: Candidate assignments stored in solution_visits table for comparison. Acceptance: When planner accepts, assignment fields updated on visits. Slinga template remains unchanged. Technical Constraints

• Import: External schedules converted to slingor with visits marked pinned=true .
• Daily expansion: Published slingor expanded into concrete visits for each date, all marked pinned.
• Optimization: Full schedule sent to optimization engine. Pinned visits stay fixed, unpinned visits can move.
• Solution storage: Candidate assignments stored in solution_visits table for comparison.
• Acceptance: When planner accepts, assignment fields updated on visits. Slinga template remains unchanged.

Hard constraints: Skills matching, no overlaps, time windows, and pinned flags are always respected by the optimization engine. Soft constraints: Continuity, travel time, workload balance, and unused hours recapture are optimized but can be overridden when necessary. Pinning validation: System ensures that when a visit is pinned, all earlier visits in the same shift that would conflict are also pinned. Freeze time: For real-time disruptions, visits that have started or are imminent remain pinned automatically (freeze time concept).

• Hard constraints: Skills matching, no overlaps, time windows, and pinned flags are always respected by the optimization engine.
• Soft constraints: Continuity, travel time, workload balance, and unused hours recapture are optimized but can be overridden when necessary.
• Pinning validation: System ensures that when a visit is pinned, all earlier visits in the same shift that would conflict are also pinned.
• Freeze time: For real-time disruptions, visits that have started or are imminent remain pinned automatically (freeze time concept).

Scalable AI stack, modern data model, and professional user interface. Built for enterprise scale while maintaining the simplicity that makes CAIRE powerful.

Enhanced Optimization Engine: Improved solver performance with better constraint handling and faster optimization cycles GraphQL API: Unified API layer replacing REST endpoints for better performance and developer experience Microservices Architecture: Modular design enabling independent scaling of optimization, analytics, and data processing Real-time Processing: Support for live schedule updates and instant optimization feedback Advanced Caching: Intelligent caching layer reducing optimization job times by up to 40% New Data Model

• Enhanced Optimization Engine: Improved solver performance with better constraint handling and faster optimization cycles
• GraphQL API: Unified API layer replacing REST endpoints for better performance and developer experience
• Microservices Architecture: Modular design enabling independent scaling of optimization, analytics, and data processing
• Real-time Processing: Support for live schedule updates and instant optimization feedback
• Advanced Caching: Intelligent caching layer reducing optimization job times by up to 40%

Normalized Schema: Complete redesign for better data integrity and query performance Slingor Support: Native support for recurring weekly patterns (slingor) as first-class entities Revision System: Built-in versioning for schedules, solutions, and optimization scenarios Flexible Constraints: Enhanced constraint model supporting complex organizational rules Audit Trail: Comprehensive logging of all schedule changes and optimization decisions Multi-tenancy: Improved tenant isolation and data security at the database level Redesigned UX

• Normalized Schema: Complete redesign for better data integrity and query performance
• Slingor Support: Native support for recurring weekly patterns (slingor) as first-class entities
• Revision System: Built-in versioning for schedules, solutions, and optimization scenarios
• Flexible Constraints: Enhanced constraint model supporting complex organizational rules
• Audit Trail: Comprehensive logging of all schedule changes and optimization decisions
• Multi-tenancy: Improved tenant isolation and data security at the database level

Professional Calendar View: Industry-leading calendar component for intuitive drag-and-drop scheduling Status-Based Visual System: Color coding by scheduling status (optional, mandatory, priority) rather than care type Real-time Validation: Instant feedback on constraint violations during drag operations Optimization Scenarios: Pre-configured optimization presets (Daily Plan, New Clients, Disruption Management) with customizable weights Comparison Mode: Side-by-side comparison of schedule revisions with delta metrics Mobile Responsive: Optimized experience across desktop, tablet, and mobile devices

• Professional Calendar View: Industry-leading calendar component for intuitive drag-and-drop scheduling
• Status-Based Visual System: Color coding by scheduling status (optional, mandatory, priority) rather than care type
• Real-time Validation: Instant feedback on constraint violations during drag operations
• Optimization Scenarios: Pre-configured optimization presets (Daily Plan, New Clients, Disruption Management) with customizable weights
• Comparison Mode: Side-by-side comparison of schedule revisions with delta metrics
• Mobile Responsive: Optimized experience across desktop, tablet, and mobile devices

Handle 10x more schedules simultaneously 50% faster optimization job completion Support for organizations with thousands and tens of thousands of employees Sub-second response times for common operations

• Handle 10x more schedules simultaneously
• 50% faster optimization job completion
• Support for organizations with thousands and tens of thousands of employees
• Sub-second response times for common operations

Intuitive drag-and-drop scheduling interface Reduced learning curve for new schedulers Better visual feedback and constraint validation Streamlined workflows for common tasks

• Intuitive drag-and-drop scheduling interface
• Reduced learning curve for new schedulers
• Better visual feedback and constraint validation
• Streamlined workflows for common tasks

How CAIRE aligns with the EU AI Act, ensuring transparency, safety, and human control.

As the European Union’s Artificial Intelligence Act comes into effect, business leaders seek clarity on how AI solutions align with regulations. At CAIRE, we prioritize transparency.

Our optimization engine is distinct from generative AI or predictive ML models. It does not learn patterns from large training datasets. Instead, it relies on pre-programmed algorithms and rules to find optimal solutions in a fully deterministic way.

Because we do not use "training data" as defined in the AI Act, many compliance obligations simply do not apply. There are no hidden biases from training sets, no opaque statistical models, and no hallucinations.

Yes. The EU AI Act defines AI systems broadly to include rule-based, logic-driven, and optimization algorithms. CAIRE qualifies because it performs reasoning and decision-making based on user input and hard-coded constraints.

However, while it fits within the scope, the deterministic, rule-based nature of our solver algorithms ensures that it avoids the Act’s strictest regulatory burdens typically reserved for more opaque or adaptive AI systems.

Every decision made by our solution can be traced back to pre-programmed rules (e.g., "minimize travel time," "ensure continuity").

This aligns perfectly with the AI Act’s transparency considerations.

• Explicit Logic: All constraints and scoring logic are explicitly configured and do not change post-deployment based on user input.
• Reproducible: Outputs are consistent. The same input always yields the same optimized schedule.
• Explainable: Users can analyze why a particular schedule was generated.

Most use cases for schedule optimization fall into operational optimization (logistics, resource planning), which are generally not considered "High Risk" under the AI Act.

However, in HR applications that significantly affect workers' rights (like shift allocation), systems can fall under the high-risk category. CAIRE addresses this by design:

• Human-in-the-Loop: CAIRE is a decision-support tool. It proposes schedules, but human planners always retain the ability to override decisions and give final approval.
• Worker Well-being: The system is configured to respect union rules, rest periods, and fair distribution of work.
• No Replacement: Our technology supports human planners; it does not replace human intuition and empathy.

Your data remains yours. No Training on Customer Data: We do not use your data to train external generative models. No Hidden Biases: Since no training data is used, there are no inherited biases from external datasets. Non-Adaptive: The system does not self-modify post-deployment.

• No Training on Customer Data: We do not use your data to train external generative models.
• No Hidden Biases: Since no training data is used, there are no inherited biases from external datasets.
• Non-Adaptive: The system does not self-modify post-deployment.

Privacy Policy → Terms of Service →

The upcoming CAIRE 2.0 platform (planned for early 2026) will maintain and enhance all existing compliance measures while introducing new capabilities:

• Enhanced Audit Trail: Comprehensive logging of all optimization decisions and data access with built-in revision system
• Improved Explainability: Better visualization of why specific schedule decisions were made, with constraint violation reporting
• GraphQL API: More granular access control with field-level permissions and improved query transparency
• Advanced Data Minimization: New v2.0 data model with better control over what data is stored and processed
• Same Compliance Foundation: All existing GDPR and AI Act compliance measures preserved and enhanced

CAIRE 2.0 maintains the same deterministic, rule-based optimization approach. The enhanced architecture provides better audit capabilities and transparency while preserving the "no black box" principle that makes CAIRE compliant today.

Learn more about CAIRE 2.0 →