XingBan AI · AI infrastructure for elderly care
Making elderly care
smarter.
XingBan AI is an AI infrastructure provider for elderly care: a model, compute, an operating system and role-based digital workers integrated into one foundation deployed on the institution's own premises. Nine roles, one digital worker each; raw data does not leave the site.
All figures are sourced; targets and measurements are kept separate.
- 9roles Digital workers
- 90items Quality monitoring metrics
- 60%+ Lower total cost of ownership
- 41.7K Care homes in China
The context
The caregiver shortfall exceeds 10 million; adding staff is no longer a viable path.
About 500,000 carers hold a qualification and their average age is 50; average occupancy is 45% against an average closure-or-transfer rate of 50% (company statistics, 2026). Workforce and cost are under pressure at the same time.
Repetitive work goes to the system; judgement stays with the people on site.
- 10M+ Caregiver shortfall in China Source: company statistics, 2026
- 500K Certified carers Source: company statistics, 2026
- 45% Average occupancy of care homes Source: company statistics, 2026
Industry figures are used to frame the problem; sources differ in definition, and forecasts are not presented as current fact.
The stack
From model to post: five layers integrated on one server.
XingBan OS for elderly care has five layers: compute, operating system, model, agent layer and digital workers. Each can be delivered on its own; together they form the foundation of an entire site.
- L5 Digital workers Nine posts, each with one digital worker, a defined remit, permissions and fixed deliverables.
- L4 Agent foundation Six agents divide the work, orchestrated outside and collaborating inside.
- L3 XingHan LLM An elderly-care-specific model, MoE architecture, deployed across device, edge and cloud.
- L2 XingBan OS A six-layer (L0 perception to L5 governance) AI-native operating system unifying people, devices, materials, environment, systems and data.
- L1 XingBan Ark Compute produces tokens, storage retains data, orchestration runs agents.
XINGHAN · Elderly-care-specific model
A model must understand elderly care to be usable on the floor.
XingHan is pre-trained on elderly-care text and fine-tuned on real nursing records. It uses a MoE architecture, deploys across device, edge and cloud, and supports private deployment with data kept inside the domain.
Explore XingHan→- MoE
- Routes each request to the relevant expert parameters.
- Device · edge · cloud
- 80% of inference stays on site (target); works offline.
- 200ms
- Target latency (lab-benchmark condition; evaluation protocol pending).
- F1 0.966
- Measured on supervised fine-tuning of care records (lab-benchmark condition).
XINGBAN ARK · Compute–storage–orchestration server
One server carries the AI load of an entire care home.
Compute produces tokens, storage retains data, orchestration runs agents. A single Ark server virtualises at least 8 logical servers, holds up to 384TB on site and reduces total cost of ownership by more than 60%. The form factor is a single machine, with dual-machine redundancy where high availability is required.
| Series | Positioning | Virtualisation | VRAM | Storage |
|---|---|---|---|---|
| X1 | Entry inference≥8 logical servers · 24–48GB VRAM · 16–32TB | |||
| X2 | Standard training≥16–24 logical servers · 96–192GB VRAM · 64–128TB | |||
| X3 | High-density compute≥32–64 logical servers · 320–640GB VRAM · 192–384TB |
Specifications follow the XingBan Ark model and marketing parameter specification (Contract Annex A, 2026); the 60%+ TCO figure is that document’s published position.
Get configuration and pricing→
Digital workers
Nine posts, one digital worker for each.
Eight are demo-ready with POC available; the inspection role is planned. Every role states its remit, its auditable metrics and its current stage.
- 01
Finance
Month-end reconciliation reduced from three days to one hour.
Demo-ready · POC available - 02
Operations
Rosters, cost and occupancy presented in one view.
Demo-ready · POC available - 03
Care
Documentation reduced by 30%; time with residents up 20%.
Demo-ready · POC available - 04
Assessment
Admission assessment reduced from three days to thirty minutes.
Demo-ready · POC available - 05
Family
Status available at any time; fewer calls to the nurses’ station.
Demo-ready · POC available - 06
Compliance
Ledgers generated to specification, without pre-inspection surges.
Demo-ready · POC available - 07
Nutrition
Dietary restrictions and nutrition constraints recorded in one system.
Demo-ready · POC available - 08
Rehabilitation
The plan adjusts weekly to the resident’s condition.
Demo-ready · POC available - 09
Inspection
Night patrols carried out by a digital worker.
Planned · 2027 Q2
Proven on site
The product is proven in our own institution before it is delivered to peers.
Shanghai Dahua Welfare Home was founded in 1999, and more than two decades of care practice form our earliest validation environment. Digital workers, smart access control and the Ark server were all validated there first; the technology, care and digitalisation entities share one building.
The first stage was care practice, the second fixed that practice into systems, and the third taught the AI that method.
-
Shanghai Dahua Welfare Home
Founded 1999, Minhang’s first private welfare home
-
Zhuanqiao Care Home
Smart care home pilot
-
Minhang Civil Affairs
Regional supervision
How we deliver
Validate in one building first, then deliver across a region.
One foundation, three deliveries. Institutional care is validated first, then extended to communities and households, with oversight last.
- 01
Care homes
41,700 care homes nationwide (Ministry of Civil Affairs, 13 Jan 2026). One Ark deploys the full stack; nine digital workers go live post by post.
Explore→ - 02
Property & home care
With large property groups, the same foundation reaches the household: health monitoring, visits and family connection. This pathway is planned; no project has yet been delivered.
Explore→ - 03
Government
90 metrics scanned daily, with findings, remediation and rechecks fully traced and a regional view by institution. No inspection outcome is promised.
Explore→
Book a demo
Deploy nine role-based digital workers on site.
One on-site demonstration of nine role-based digital workers running in real scenarios. A solution consultant will call to confirm the time and the focus of the demonstration.