live · matching pool
—
護理員候選 · providers ready
AI

Pick a user from the panel below — I'll suggest weights based on their care history.

Knowledge graph + vectors + LLM

Care matching, reimagined.

Five reasoning paths. Three databases. One score. Built for operators who match elders with the right caregiver, every time.

— elders
— caregivers
— graph edges
edges · neo4j
SERVED_BY · HAS_COND · LOCATED_IN
01
Hard Filter
cert · geo · budget
02
Vector Retrieval
milvus · 1024-d
03
KG Retrieval
neo4j · 5 paths
04
Score Fusion
vector ⊕ kg
05
LLM Re-rank
deepseek-v4-flash
configure match

New match request

—
auto-normalised to 1.00
history .30
condition .15
skill .40
collab .10
geo .05
what do these weights mean? ›
  • Skill — Does the caregiver's qualification fit the service type? (Rehab → PT/OTA/PTA, Nursing → RN, etc.)
  • History — This caregiver's own service-record strength for this kind of case — strongest when they've served this elder before.
  • Condition — How well the caregiver's experience fits the elder's medical conditions (dementia, Parkinson's, wheelchair use, etc.).
  • Collab — A "similar elders" signal — caregivers who worked out well for elders like this one score higher, even with no direct history here.
  • Geo — Geographic proximity — same or nearby district.
est. 3–4 min · cost ~$0.005