Work/2026
Qhali Ayllu
Multimodal health triage for rural Peru — anemia and tuberculosis screening with nothing but a smartphone.
1stof 9 teams, MIT GTL AI
2diseases screened offline
0lab equipment required
A multimodal Android app for early anemia and tuberculosis screening in rural Peru using only a smartphone — no internet or lab equipment required. Selected as the top project among 9 teams at the MIT Global Teaching Labs AI Workshop (MIT MISTI × UTEC) and pitched to MIT instructors and industry professionals.
Highlights
- Two-stage computer vision. EfficientNetB0 segments the conjunctiva, then DenseNet121 classifies pallor associated with anemia — a pipeline tuned for phone cameras in uncontrolled lighting.
- Audio as a second modality. Cough recordings are processed with Librosa (MFCC, zero-crossing rate) and fed to SVM, logistic regression and decision trees — models chosen deliberately for safe, interpretable clinical triage.
- Built for the edge case that matters. Everything runs offline on modest Android hardware, because the communities that need screening most are the ones without connectivity.
