A blood donation network that finds the nearest donor in seconds

The challenge

When somebody needs blood urgently, the search is a relative ringing everyone they know and messages forwarded through group chats until they reach someone who might help. It works, eventually. Eventually is the problem. The replacement had to reach compatible donors in seconds, respect medical eligibility rules, and never expose a donor's location beyond the person they chose to help.

What we did

1. 01Split the system into two services that never call each other — identity owns accounts and sessions, platform owns donors, requests and matching — joined only by a shared JWT secret and a one-way Redis event stream
2. 02Built matching on blood compatibility rather than type equality, filtered by availability, the 56-day donation cooldown, the search radius and each donor's own maximum travel distance, ranked nearest-first then most reliable
3. 03Escalated unanswered requests automatically: a Celery beat job doubles the radius through 10, 20 and 40 km, capped at 50, notifying the donors earlier waves missed
4. 04Kept access tokens out of the browser entirely — a BFF proxy attaches them from an httpOnly cookie, so a script injection has nothing to steal
5. 05Shipped live donor location that is visible only to the requester, only while the donation is open, and never automatically
6. 06Covered the two failure modes unit tests structurally cannot reach — a token crossing services, and an event crossing Redis — with 31 smoke checks against running containers