01
Detect
MedX agents inspect prescription context and surface medication-safety findings for the treating doctor.
Research prototype · federated medication safety
Medicine learns.
Patients stay private.
MedX lets hospitals learn from each other’s medication-safety experience without moving a single patient record. Clinicians stay in control at every step.
Federated · clinical · compute
The problem
Medication-safety knowledge stays inside hospital walls.
Each hospital sees different patients, risks, and near misses. MedX turns those isolated experiences into shared intelligence — while every patient record stays exactly where it is.
How MedX Works
From local care to shared intelligence.
02
Review
Doctors and pharmacists remain in control of every clinically meaningful action, including every override.
03
Learn Locally
Each hospital turns its own reviews, holds, and near misses into structured local learning evidence.
04
Verify
Sentinel evaluates every model update for trust and safety signals before it can influence the network.
05
Federate
Trusted updates are aggregated through Flower — only the model update leaves the hospital boundary.
06
Improve
Other hospitals benefit from the shared model update, never from another hospital's patient records.
Sentinel
Trust nothing blindly.
Every federated update passes through Sentinel’s trust and safety evaluation before it can influence the shared model.
Privacy protects the data.
Sentinel protects the learning.
Accept
Trusted
High-trust updates are aggregated in full for this round.
Downweight
Reduced influence
Suspicious but usable updates contribute with a reduced effective weight.
Quarantine
Excluded
Unsafe or anomalous updates are excluded before aggregation proceeds.
Network
One hospital learns.The network benefits.
MedX coordinates local review, Sentinel verification, and Flower-based federated aggregation — patient records never leave the hospital that holds them.
Zero patient records exchanged.
Hospital A
Local doctor & pharmacist review
Sentinel
Trust and safety verification
Federated Intelligence
Flower-aggregated model update
Hospital B
Improved local model, same isolation
Illustrative overview · see the live Federation Command Center for real experiment data
Proof
Built to be inspected.
Human Override
Every clinical decision stays with a clinician, with a recorded reason.
Evidence Lineage
Trace where local learning evidence came from, inside its hospital boundary.
Replay
Saved, verified proof runs can be replayed and inspected on demand.
Enter the prototype
See the network learn.
Enter as a seeded research identity and follow one prescription through doctor review, pharmacist safety, local learning, and cross-hospital evidence.
Enter Secure Workspace Access
Seeded research access · not production authentication