Home How it works

How it works

A model goes in. A verdict comes out.

Every model runs on the low-cost phones people actually use, under the conditions they use them in. Models that fail are optimised and tested again.

1
Intake

A team submits a model

With a test plan and sample data, from LUMS, the partner nodes or any team in a low- or middle-income country. The testbed is remotely accessible.

2
Testbed

Runs on real low-cost phones

Two instrumented racks of phones and edge hardware, under controlled:

HeatHumidityWi-Fi & 4GPower cutsMemory pressure
3
Verdict

One of three verdicts

  • Field-Ready
  • Field-Ready with Constraints
  • Not Yet Field-Ready

Plus a feasibility score, a per-device budget and a diagnosis of where it fails.

4
Field pilots

Certified tools go to the field

Field results are checked against lab predictions, so the testbed gets more accurate over time.

Optimisation loopNot yet ready? Quantisation, pruning and distillation, or a domain research track, then re-certified. Low-resource speech and language models get their own workstream.
First use cases
Pregnancy risk screening Family-planning counselling Early-grade reading
Every test produces open evidence
Device × workload mapwhich phones run which models
Optimisation resultswhich techniques deliver what
Optimised model poolwith known limits
Deployment blueprintsopen-source by Year 4