Thanks to visit codestin.com
Credit goes to github.com

Skip to content

SRS.md

Latest commit

 

History

History
337 lines (258 loc) · 14.1 KB

SRS.md

File metadata and controls

337 lines (258 loc) · 14.1 KB

Software Requirements Specification (SRS)

Project Trij — FieldMed Triage System

Version: 1.0 Date: May 12, 2026 Prepared for: Gemma 4 Good Hackathon (Kaggle + Google DeepMind) Track: Health & Sciences / Global Resilience

1. Introduction

1.1 Purpose

Trij is a progressive web application that enables community health workers (CHWs) in remote, low-connectivity areas to perform AI-assisted medical triage using Google DeepMind's Gemma 4 models. It provides preliminary assessment of wounds, rashes, and medical documents through multimodal AI — entirely on-device, preserving patient privacy.

1.2 Scope

The system covers:

  • Image-based triage assessment (wounds, rashes, skin conditions)
  • Medical document analysis (lab reports, referral letters)
  • Voice-guided assessment workflows in local languages
  • Offline-first architecture with background sync
  • Patient record management
  • Confidence-scored recommendations with referral guidance

1.3 Definitions

Term Definition
CHW Community Health Worker (primary user)
Triage Preliminary assessment of medical urgency
Gemma 4 Google DeepMind's open multimodal LLM family
PWA Progressive Web Application
E2B/E4B Gemma 4 edge-optimized model variants (2B / 4B params)
RAG Retrieval-Augmented Generation

2. User Roles

2.1 Community Health Worker (CHW)

  • Conducts triage assessments
  • Views patient history
  • Manages offline patient records
  • Receives referral recommendations

2.2 Supervisor / Clinic Admin

  • Reviews CHW assessments
  • Syncs data from field workers
  • Manages referral outcomes
  • Views analytics dashboard

2.3 Patient

  • Minimal interaction (consent-based data subject)
  • Receives referral slips (via CHW)

3. Functional Requirements

FR-01: Image-Based Triage Assessment

Item Description
Input Photo of wound, rash, or skin condition (phone camera or gallery)
Process Gemma 4 vision analyzes image; model generates structured JSON assessment
Output Condition classification, urgency level (Green/Yellow/Red), confidence score, possible conditions list
Constraints Must work offline; inference runs on-device via E2B/E4B

FR-02: Medical Document Analysis

Item Description
Input Photo of lab report, referral letter, or prescription
Process Gemma 4 OCR + reasoning extracts key metrics and flags abnormalities
Output Structured summary of findings, abnormal values highlighted, plain-language explanation
Constraints Must handle handwritten text and low-quality photos

FR-03: Voice-Guided Assessment

Item Description
Input Spoken answers in local language (Web Speech API)
Process Gemma 4 processes audio transcription, generates follow-up questions via function calling
Output Dynamic question flow adapted to patient responses
Constraints Must support 140+ languages; works offline with cached language packs

FR-04: Patient Record Management

Item Description
Create Register patient with minimal fields (ID, age, sex, location)
Read View patient history across visits
Update Add new assessments to patient timeline
Constraints All data stored locally via IndexedDB; sync to Supabase when online

FR-05: Offline-First Operation

Item Description
Triage Full assessment capability without internet
Storage All records cached in IndexedDB
Sync Background sync to Supabase when connectivity returns
Conflict Last-write-wins with server timestamp reconciliation

FR-06: Confidence Scoring & Explanation

Item Description
Score Every assessment includes 0-100 confidence score
Rationale Brief explanation of key visual features used for assessment
Low-Conf Auto-suggests "Refer to clinic" when confidence < 70

FR-07: Referral Management

Item Description
Generate Auto-generate referral slip PDF (image + assessment summary + QR)
Track Mark referral as "sent", "received", "resolved"
Notify Alert supervisor when new referrals created (when online)

FR-08: Multilingual Interface

Item Description
UI App UI available in English, French, Swahili, Hindi, Portuguese, Arabic
Assessment Voice and text I/O in user's chosen language
Reports Generated reports in local language + English

FR-09: Supervisor Dashboard

Item Description
Map Geolocated view of assessments in region
Queue Pending referral queue
Analytics Assessment volume, common conditions, sync status
Export CSV export of anonymized assessment data

4. Non-Functional Requirements

NFR-01: Performance

Metric Target
Image inference time < 10 seconds (E2B quantized)
App cold start < 3 seconds
Voice response latency < 2 seconds
Sync time (100 records) < 30 seconds (on 3G)

NFR-02: Privacy & Security

  • All PHI (Protected Health Information) stays on-device until explicit sync
  • End-to-end encryption for synced data (Supabase row-level security)
  • No patient data sent to external AI APIs
  • Gemma 4 runs locally — no cloud inference
  • HIPAA-compliant data handling patterns

NFR-03: Reliability

  • 100% functionality offline (triage, records, voice)
  • Graceful degradation when storage exceeds 80%
  • Automatic sync queue with retry + exponential backoff

NFR-04: Compatibility

  • Chrome / Firefox / Safari (latest 2 versions)
  • Android Chrome (PWA installable)
  • iOS Safari (PWA installable)
  • Minimum 4GB RAM device
  • Camera resolution ≥ 8MP

NFR-05: Scalability

  • Support up to 10,000 patients per deployment instance
  • Support up to 500 CHWs per deployment region
  • Supabase handles backend scaling

5. System Architecture

┌─────────────────────────────────────────────────────┐
│                   Client (Browser)                    │
│  ┌──────────┐  ┌──────────┐  ┌──────────────────┐   │
│  │  React   │  │  WebLLM  │  │  Web Speech API  │   │
│  │  (Vite)  │  │ (Gemma4) │  │  (Voice I/O)     │   │
│  └────┬─────┘  └────┬─────┘  └────────┬─────────┘   │
│       │             │                  │              │
│  ┌────▼─────────────▼──────────────────▼──────────┐  │
│  │           Service Worker Layer                   │  │
│  │  ┌─────────────┐  ┌────────────────────────┐   │  │
│  │  │  Cache API  │  │  Background Sync API   │   │  │
│  │  └─────────────┘  └───────────┬────────────┘   │  │
│  └────────────────────────────────┬────────────────┘  │
└───────────────────────────────────┼───────────────────┘
                                    │
            Online ┌────────────────┴────────────────┐
                   │                                 │
            ┌──────▼──────┐              ┌───────────▼───┐
            │  Supabase   │              │  IndexedDB    │
            │  (Cloud)    │◄────────────►│  (Local)      │
            │  - Auth     │   Sync Job   │  - Patients   │
            │  - Postgres │              │  - Assessments│
            │  - Storage  │              │  - PendingSync│
            │  - Edge Fns │              └───────────────┘
            └─────────────┘

6. Data Model

Patients

{
  "id": "uuid",
  "chw_id": "uuid",
  "identifier": "string (local ID)",
  "age_years": "number",
  "sex": "enum('M','F','other')",
  "location": "geojson point",
  "created_at": "timestamp",
  "updated_at": "timestamp",
  "last_sync": "timestamp"
}

Assessments

{
  "id": "uuid",
  "patient_id": "uuid",
  "chw_id": "uuid",
  "images": ["string (base64 stored locally, URL when synced)"],
  "condition": "string",
  "confidence": "number (0-100)",
  "urgency": "enum('green','yellow','red')",
  "possible_conditions": [{ "name": "string", "probability": "number" }],
  "recommendation": "string",
  "voice_log": "string (transcript)",
  "language": "string",
  "referral_status": "enum('none','sent','received','resolved')",
  "created_at": "timestamp",
  "created_at_offline": "timestamp",
  "synced_at": "timestamp"
}

CHW Profiles

{
  "id": "uuid",
  "name": "string",
  "language": "string",
  "region": "string",
  "supervisor_id": "uuid",
  "device_id": "string",
  "last_sync": "timestamp"
}

7. API Endpoints (Supabase Edge Functions)

Endpoint Method Purpose
/sync/assessments POST Batch upload assessments from offline queue
/sync/patients POST Batch upload new patients
/sync/pull GET Pull latest references, protocols, language packs
/referrals GET Supervisor view of referral queue
/analytics/region GET Aggregated analytics for supervisor dashboard
/ai/download-model GET Download latest Gemma 4 E2B model shard
/ai/language-pack GET Download voice/language pack for offline use

8. UI/UX Requirements

8.1 Screen Map

  1. Login — PIN-based (offline) + Supabase Auth (online)
  2. Dashboard — Recent patients, quick triage button, sync status indicator
  3. New Triage — Camera capture → analysis → result → patient linkage
  4. Patient Detail — Visit timeline, condition history
  5. Document Scanner — Photo capture → OCR → structured output
  6. Referral View — Generated referral slip, status tracker
  7. Supervisor Dashboard — Map, queue, analytics

8.2 Design Principles

  • Large touch targets (rural users, varying tech literacy)
  • High-contrast mode for outdoor use
  • Minimal text input (voice + photo preferred)
  • Single-thumb navigation
  • Offline indicator always visible
  • Language toggle always accessible

9. Assumptions & Constraints

  • CHW devices have ≥ 4GB RAM and a camera
  • Internet connectivity is intermittent (not always available)
  • CHWs have basic literacy and smartphone familiarity
  • Gemma 4 E2B/E4B models can be quantized to run in-browser via WebLLM
  • Supabase free tier sufficient for pilot deployment
  • Patient consent is obtained verbally per local practice

10. Glossary

Term Meaning
CHW Community Health Worker
PHI Protected Health Information
PWA Progressive Web Application
RAG Retrieval-Augmented Generation
MoE Mixture of Experts (model architecture)
RLS Row-Level Security (Supabase)

This SRS is a living document. Updates will be tracked via git commits.