MEDIREACH

LOGIN SCREEN
CONSENT SCREEN
AGE SCREEN
SYMPTOMS SCREEN
OPTIONAL VITALS SCREEN
DURATION SCREEN
CONFIRMATION SCREEN
FINAL REPORT GREEN
FINAL REPORT AMBER
DOSAGE RECOMENDATION
CHW QR CODE
FINAL REPORT RED
CHW DASHBOARD ABOVE
CHW DASHBOARD BELOW
SUPERVISOR DASHBOARD
SUPERVISOR WORKER ACCESS

Inspiration

Every year, over 4.5 billion people lack access to essential health services (WHO, 2023). But in many low-resource settings, the primary failure is not infrastructure — it is delay in decision-making.

A Community Health Worker (CHW) often faces a critical question:

"Is this safe to manage at home, or is this an emergency?"

This hesitation — driven by uncertainty — leads to delayed care and preventable deaths.

We built MediReach v3 to eliminate this hesitation gap by giving frontline workers a fast, reliable, and explainable triage system that works anywhere, even offline.

What it does

MediReach v3 is an offline-first clinical triage system designed for Community Health Workers.

In under 60 seconds, it:

Classifies patient condition into RED / AMBER / GREEN
Calculates an urgency score (0–100)
Provides plain-language care guidance
Works offline on any smartphone
Supports multilingual voice input (Tamil, Hindi, Swahili, English)
Shows nearest clinics using GPS
Feeds data into a real-time epidemiological heatmap

It does not replace doctors — it ensures patients reach them at the right time.

How we built it

We engineered a two-layer safety architecture:

1. Deterministic Clinical Engine

Based on WHO IMCI protocols
Runs fully on-device (IndexedDB + Service Worker)
Executes in < 50 ms
Produces severity + urgency score:

$$ U = U_{\text{base}}(S) + \Delta_{\text{duration}} + \Delta_{\text{category}} + \Delta_{\text{compound}} $$

This layer guarantees safety.

2. AI Enrichment Layer

Uses Llama 3 70B via Groq
Converts rule outputs into:
- Human-readable explanations
- Local-language instructions
- Family communication guidance

AI cannot override clinical severity.

Supporting Systems

Offline-first storage: IndexedDB → Supabase sync → localStorage fallback
Secure access: QR login + JWT (HttpOnly)
Voice parser: multilingual symptom detection
Responsive UI: mobile (CHW) + desktop (supervisor dashboard)
Epi-Heatmap: real-time outbreak pattern detection

Challenges we ran into

1. Offline reliability vs real-time intelligence

Balancing a fully offline system with optional AI required strict separation between deterministic safety logic and AI enrichment.

2. Service Worker cache inconsistency

App updates caused stale UI crashes in offline mode.
We solved this using navigation fallback + cache revalidation strategy.

3. Multilingual voice ambiguity

Different languages describe symptoms differently.
We built a custom keyword-mapping parser instead of relying on generic speech models.

4. Security without friction

Traditional login systems fail in field environments.
We designed a QR-based access system that is both secure and fast.

Accomplishments that we're proud of

Built a fully offline clinical triage engine
Integrated WHO-aligned medical logic into code
Achieved sub-60-second decision flow
Developed multilingual voice triage
Created a real-time epidemiological heatmap
Designed a system that scales across countries without core changes
Implemented automated testing for clinical rules

What we learned

Safety > intelligence in healthcare systems
AI is most powerful as a translator, not a decision-maker
Offline-first design is not optional in global health — it is foundational
Explainability increases trust and action
Real-world constraints (language, connectivity, device limits) define success more than algorithms