Projects

Background

Rural clinics face triage bottlenecks, weak connectivity, and fragmented follow-up across visits.

Methods

• Designed WhatsApp-first intake + triage flows with USSD fallback paths.
• Modeled privacy-first consent + referral handoffs for constrained settings.

Findings

• Reduced friction in intake workflows and preserved continuity under low bandwidth.
• Established ethical logging + consent language that survives constrained UX.

Next

Build clinician-facing dashboard + evaluate usability with real clinic partners.

Background

Quality-of-life instruments can behave differently across groups, risking biased interpretations and invalid comparisons.

Methods

• Ran CFA-based invariance testing across demographic groupings.
• Screened for DIF and documented thresholds where comparisons become unreliable.

Findings

• Identified sensitivity points where measurement structure shifts across groups.
• Produced interpretation guidance and candidate revisions for improved fairness.

Next

Finalize reporting + propose recalibration strategy with stakeholders.

Background

Risk prediction can be improved by encoding positional/biological structure into features rather than treating variants as flat inputs.

Methods

• Engineered position-weighted features that preserve genetic structure.
• Built predictive pipeline and validated against synthetic population scenarios.

Findings

• Demonstrated that structure-aware feature engineering can improve signal capture.
• Produced a reusable pipeline design for future real-world datasets.

Next

Adapt pipeline to real cohort data and evaluate generalization.

Background

Feasibility depends on where need intersects with connectivity and practical workflow constraints.

Methods

• Built modelling + visualization to estimate telehealth feasibility under constraints.
• Designed USSD-based triage concept to cover low-connectivity settings.

Findings

• Mapped feasibility bottlenecks and workflow choke points.
• Produced an actionable architecture for low-bandwidth triage + routing.

Next

Package models into a small public dashboard + policy-facing narrative.

Background

Sickle-cell trait protection varies across contexts; parasite genetics may modulate outcomes and explain heterogeneous findings.

Methods

• Synthesizing genotype-specific evidence across malaria studies.
• Formulating testable hypotheses for parasite-mediated protection mechanisms.

Findings

• Early synthesis suggests plausible genotype-driven interaction effects worth formal testing.

Next

Move from synthesis → preregistered analysis plan + target dataset options.

Background

Newcomers face a mismatch between local hiring norms and their lived strengths; curriculum must translate capability into employer-recognized signals.

Methods

• Designed 12-week program with behaviorally specific skill ladders.
• Built program logic model and implementation plan from stakeholder + evidence inputs.

Findings

• Produced a scalable curriculum structure aligned to employer-valued behaviors.
• Created artifacts that support delivery consistency across cohorts.

Next

Iterate via participant feedback + add portfolio-based assessment components.