When a donor can’t match their intended recipient, KindRed finds another possibility.
KindRed connects living donor–recipient pairs across transplant programs, helping teams discover exchange possibilities, coordinate review, and recover when a proposed pathway changes.
- KidneyPaired donation
- LiverPaired donation
Three families. Each has a donor who can't give to their own patient.
KindRed finds a cycle: each donor gives to the next family's patient.
Then a lab test fails, and the loop comes undone.
KindRed finds a family nearby. A coordinator approves the change.
Most of these families never get there.
India does more living-donor transplants than any other country, yet it meets only about 7% of its yearly need for kidneys and about 1.5% for livers. Many willing family donors are ruled out by blood group alone. Swaps would help them, but India has managed 1,839 kidney swaps in 24 years and 259 liver swaps at 7 centres. Finding a match is the easy part. The swap falls apart afterwards, when a late test fails and a coordinator has to start again by phone.
Impact, in a simulated year of a national kidney pool
76%
more transplants
when hospitals share one pool instead of each matching alone
14%
fewer wasted lab tests
than a 3-way matcher that ignores failed tests, with no fewer transplants
~25
broken swaps repaired a year
instead of starting again from scratch
0.7
weeks sooner
to a match that actually holds
20 simulated years of about 540 families each, measured by us. The generator reproduces published match rates (Saidman et al. 2006).
How a swap works
Willing, but the wrong match.
Each family has someone ready to donate, but not to their own patient. The blood group is wrong, or the patient carries antibodies that would reject the kidney.
One pool, many hospitals.
One hospital rarely has the right families at the same time. KindRed shares only the medical facts matching needs across hospitals. Names stay with each family's own hospital.
The loop closes.
Each donor gives to the next family's patient. Three families, three kidneys that fit, and nobody donates unless their own patient receives one too.
2-way or 3-way?
A family whose donor can’t give to their own patient joins the pool. A swap is a loop of families where each donor gives to the next family’s patient, so every patient receives and nobody gives without their own family getting one back.
2-way swap
Family A's donor gives to B's patient, and B's donor gives to A's patient. Simple, and the most common.
In the graph: a donor-to-patient link and one straight back (1 hop out, 1 back).
3-way swap
A gives to B, B gives to C, C gives back to A. It helps the families nobody can swap with directly: the hard-to-match patients.
In the graph: a loop of 3 links. A shallow, 2-way search can't see it; only a deeper search does.
Every link is tested before surgery (a lab crossmatch for kidney, a donor work-up for liver). If one fails, the swap breaks. KindRed keeps a 2-way swap inside each 3-way one as a fallback, and searches nearby for a replacement.
Someone watching over every family, week after week.
KindRed’s agents follow each patient from a falling lab result to a transplant. Every step says who did it, what it looked at first, and why.
1
Notices when someone gets worse
Reads every new lab result and compares it with medical guidelines and with what it remembers. It speaks up only when a patient gets worse, so nobody is buried in alerts.
Health watch
2
Takes the next step, every week
Looks at the patient's situation and does what's due: search for a swap, run the tests, ask for papers, book the surgery. It shows what it looked at and why.
Coordinator
3
Mends a swap when a test fails
When one donation can't go ahead, it looks only at the families nearby for a replacement, keeps a backup ready, and leaves every other swap untouched.
Matching and repair
4
Keeps everyone in the loop
Books appointments, chases missing papers, and tells each family what is happening, in plain words and their own language.
Scheduler, clerk, family liaison
A person approves every repair. The AI can only use checked tools, never edits the records directly, and every family it mentions is confirmed first. All families here are synthetic; KindRed is a prototype, not a clinical tool.
Why it needs a graph
Families, donations and memory, all connected.
A swap is a loop
Families are points and possible donations are the lines between them. A swap is a loop of lines, and finding loops is what a graph database does best.
Loops found by traversal in FalkorDB
Looking further finds more
A direct trade between two families is easy to spot. A three-family loop helps the patients nobody can trade with directly, and it only shows up when you look one step further.
2-way and 3-way searches, side by side
It remembers, and shows why
Every test, decision and conversation is kept in the same graph. Each agent picks up where another left off, and every answer comes with the path behind it.
Memory stored as connected nodes
Follow a family to a transplant.
Meet Sunita, who needs a kidney, or Rajesh, who needs part of a liver. Move forward one week at a time and watch the agents find a swap, run the tests and book the surgery. Stuck anywhere? Ask Kin, at the bottom of every screen.