Features — JemmaPass AI Health Companion

✓ DEMO Fully implemented & running in the contest APK ◐ PARTIAL Code present, limited scope in demo ○ ROADMAP Planned for production — stub UI only in demo

🧠 AI Engine — Gemma 4 via LiteRT

✓ DEMO

Gemma 4 On-Device Chat

Full conversational AI companion via LiteRT-LM (E4B 3.4 GB or E2B 2.4 GB). Multi-turn context, system prompts, patient-aware reasoning.

GemmaSessionLiteRT-LME4B / E2B

✓ DEMO

68 Gemma @Tool Functions

Gemma 4 function-calling into the clinical KB: resolveDrug, checkDdi, checkOneDrugAgainstFocusProfile, resolveAllergy, searchCodes, triggerRedAlert and 10 more.

JemmaTools.ktToolSet68 @Tools

✓ DEMO

Med-Scan AI Pipeline

Camera → ML Kit OCR → structured drug extraction pipeline powered by Gemma. GemmaMedScanPipeline with MedScanTools passes OCR text through Gemma for structured drug entry.

GemmaMedScanPipelineML Kit OCR

✓ DEMO

Voice Input (STT)

Speech-to-text module fully implemented (ai/stt/). Wired to the chat interface — speak to Jemma to add entries, ask clinical questions, or give commands. Works on all Android devices with a microphone.

ai/sttAndroid SpeechRecognizer

✓ DEMO

Text-to-Speech (TTS)

TTS module fully implemented (ai/tts/). Jemma speaks her responses aloud — clinical summaries, drug interaction alerts, and plain-language explanations. Hands-free mode for emergency scenarios.

ai/ttsAndroid TTS

○ ROADMAP

Multilingual UI (FR / JA)

Gemma generates patient summaries in English, French, and Japanese via its KB. Full UI localization (strings for all 3 locales) planned post-contest.

IpsTranslationsRepositoryFR / EN / JA

🛡️ Clinical Knowledge Base & Safety Shield

✓ DEMO

3.36 GB Clinical KB — DDInter 2.0

Full on-device SQLite FTS5 knowledge base: 302,516 DDI records, 8,359 drug-disease mappings, SNOMED CT, RxNorm, ATC, LOINC, FoodData. Sub-50ms queries.

KnowledgeBaseServiceSQLite FTS5DDInter 2.0

✓ DEMO

DDI + Allergy + Drug-Disease Cross-check

KbCrossCheck scans a drug against all three pillars at once: allergy collisions (SNOMED + ATC class), DDI (Major/Moderate), drug-disease contraindications.

KbCrossCheck.ktCrossCheckResult

✓ DEMO

Red Alert UI Trigger

When Gemma detects a critical collision, it calls triggerRedAlert which fires a full-screen alert with sound on the device. Fully event-driven via JemmaToolEvent.

JemmaToolEventRedAlert

📋 IPS Clinical Pillars — 18 total

✓ DEMO — Fully editable

3 Active Pillars

📝 Patient — demographics, blood type, address, contacts

⚠️ Allergies & Intolerances — SNOMED CT coded, criticality, manifestation

💊 Medications — ATC/RxNorm coded, dose, route, timing

○ ROADMAP — Stub UI, FHIR-modelled

14 Passive Pillars

📜 Past Problems | 💉 Immunizations | 🏥 Procedures

📱 Devices | ♿ Functional Status | 🤰 Pregnancy

🧪 Lab Results | 📜 Advance Directives | ✍️ Consents

🎯 Care Goals | 🚑 Encounters | 💼 Occupational

👨‍⚕️ Care Providers | 👥 Emergency Contacts

✓ DEMO

FHIR R4 IPS Bundle Export

JemmaFhirBundleBuilder generates a fully-compliant HL7 IPS FHIR R4 document bundle from the active pillars. Serialized via FhirR4Json.

JemmaFhirBundleBuilderHL7 IPSFHIR R4

📲 QR & Data Transfer

✓ DEMO

Triple-Layer QR Code

JemmaPayloadCodec + JemmaPayloadPruner compress the FHIR bundle ~70% via clinical pruning + GZIP. JemmaQrBitmapEncoder generates the QR. Split into multi-frame for large payloads.

JemmaPayloadPrunerJemmaQrFrameSplitterGZIP

✓ DEMO

Human-readable IPS Text

JemmaTextPayloadBuilder generates a plain-text human-readable IPS summary embedded as a QR layer, readable by any paramedic without an app.

JemmaTextPayloadBuilderPlain-text IPS

◐ PARTIAL

Live QR Scan (Camera)

Live scan module (ai/livescan/) implemented for the rescuer flow. Reads another device's JemmaPass QR and reconstructs the clinical profile in the rescue context.

ai/livescanML Kit Barcode

📡 Nearby SOS Mesh & Rescue Radar

✓ DEMO

SOS Broadcast (Victim Mode)

JemmaSosService (69 KB) + JemmaSosChunkCodec: encrypts and broadcasts the clinical profile in 131-byte chunks over Google Nearby Connections P2P_CLUSTER. Zero infrastructure.

JemmaSosServiceNearby P2PChunkCodec

✓ DEMO

Rescue Radar (Rescuer Mode)

RadarController (40 KB) + JemmaNearbySosService reassemble victim chunks, reconstruct full clinical profiles, and display victims on a 360° radar with distance estimate.

RadarControllerJemmaNearbySosService

✓ DEMO

SALT Triage Protocol

SaltCode + StatusResolver + StatusEvent: rescuers tag victims Green/Yellow/Red/Black. Tags propagate across all Nearby peers in real-time. JemmaRescuerBroadcaster handles relay.

triage/SaltCodeStatusResolverMesh Relay

✓ DEMO

BLE Fallback (Victim Discovery)

JemmaSosBleAdvertiser + JemmaSosBleScanner: BLE advertising fallback for when Nearby is not available. Ensures victim discovery even in degraded RF environments.

JemmaSosBleAdvertiserJemmaSosBleScanner

✓ DEMO

Emergency Home Screen Widget

JemmaEmergencyWidget + JemmaWidgetEmergencyService: Android home screen widget that displays blood type, critical allergies, and emergency contact — no app launch needed.

JemmaEmergencyWidgetAppWidget

○ ROADMAP

Mesh Relay & Multi-hop

mesh/relay/ module scaffolded. Multi-hop message relay (victim → relay device → rescuer) for larger disaster zones. Planned for v3.0 production.

mesh/relaymesh/transport

100% Offline & Air-Gap

JemmaPass and its Gemma 4 engine work with zero network access. All inference, clinical reasoning and data storage happen locally — no sync, no server, no outage. Perfect for disaster zones, rural areas, or simply protecting your privacy.

LiteRT on-deviceSQLiteAir-Gap

Clinical Knowledge Base

A 3.36 GB on-device knowledge base (SNOMED CT, RxNorm, ATC, LOINC, FoodData) powers JemmaPass's Clinical Shield. Gemma 4 uses native Function Calling to query it and detect drug-drug (DDI) and drug-food (DFI) interactions — all without a single byte of cloud traffic.

Gemma 4 @ToolsSNOMED CTDDI/DFI

Triple-Layer QR

JemmaPass's Clinical Forge compresses your FHIR R4 bundle by 70% and encodes it in three QR layers: binary FHIR (machine-to-machine), Gemma 4-generated human-readable IPS summary, and a locale-specific plain-text fallback readable by any generic scanner.

FHIR R4GZIP−70% payload

Multilingual AI Summaries

Gemma 4's autonomous localized reasoning generates patient summaries in the doctor's language on the fly — French, English, and Japanese.
Demo: 3 languages supported.

Gemma 4 reasoningFR / EN / JA

Privacy by Design

Your data and your AI companion never leave your device. Gemma 4 runs locally via LiteRT — zero telemetry, zero analytics, zero account required. Even your AI conversations are private.

GDPRLocal-only AIZero telemetry

Nearby · SOS Mesh

Broadcast your clinical profile over Google Nearby Mesh. No network, no QR display needed. Perfect when pinned under rubble or hands-free. BLE fallback ensures victim discovery even in degraded RF environments.

Nearby MeshP2P ClusterBLE Fallback

📋 Clinical Standard

Built on HL7 IPS —
the global standard

JemmaPass implements the International Patient Summary (IPS) standard, mandated by the EU Health Data Space and adopted by WHO. Every field maps to a FHIR R4 resource built with the Open Health Stack Kotlin FHIR library (dev.ohs.fhir:fhir-model:1.0.0-beta03, Apache 2.0 — launched by Google, now stewarded by the OHS community). Jemma's AI companion guides you through filling each section via natural conversation.

✓ AllergyIntolerance — AllergyIntoleranceBuilder

✓ MedicationStatement — MedicationStatementBuilder

✓ Patient — PatientBuilder

○ 15 Pillars — FHIR-modelled, UI planned for production

// JemmaFhirBundleBuilder.kt
// Google FHIR Kotlin SDK — dev.ohs.fhir v2.2.16.14

val bundle = Bundle.newBuilder()
  .setType(BundleTypeCode.Value.DOCUMENT)
  .addEntry(
    Bundle.Entry.newBuilder()
      .setResource(Resource.newBuilder()
        .setAllergyIntolerance(
          AllergyIntolerance.newBuilder()
            .setCode(CodeableConcept.newBuilder()
              .addCoding(Coding.newBuilder()
                .setSystem(Uri.of("http://snomed.info/sct"))
                .setCode(Code.of("372687004"))
                .setDisplay("Penicillin G")))
            .setCriticality(
              AllergyIntoleranceCriticalityCode
                .Value.HIGH))
      ).build()
  ).build()

🛡️ Interaction Shield · DDInter 2.0

The guardian that
never sleeps

JemmaPass embeds the entire DDInter 2.0 professional clinical database. With over 302,516 drug-drug interactions and 8,359 drug-disease mappings, it detects lethal combinations at the point of care. Powered by molecular identity (SMILES/InChI), Jemma's Shield operates with hospital-grade accuracy, 100% offline.

DDInter 2.0 302k+ Records SMILES reasoning Contraindication Check

🛡️ JEMMA SHIELD v4.0

[SCANNING] HL7 IPS Bundle...
> 33,934 drugs indexed.
> 6,835 conditions indexed.

MEDICATION DETECTED

Abacavir (Ziagen)

InChI=1S/C14H18N6O/c15-14-18-12-11(13(19-14)17-10-3-4-10)16-7-20(12)8-1-2-9(8)6-21...

CONDITION DETECTED

Severe Hepatic Impairment

SNOMED CT · 235856003

⚠️ CRITICAL INTERACTION LEVEL 3

Abacavir is contraindicated in patients with moderate to severe hepatic impairment. Risk of fatal lactic acidosis.

🧠 Gemma 4 · Function Calling

Ask Jemma anything —
she finds the right answer

Jemma's brain is Gemma 4 running via LiteRT, augmented by native @Tool function calling into a 3.36 GB clinical knowledge base (SNOMED CT, RxNorm, ATC, LOINC). Say "is sashimi safe with my Warfarin?" — Jemma reasons over your medications, queries the DFI engine and gives you a clinical-grade answer. Entirely offline. Under 2 seconds.

Google FHIR Kotlin SDK Android SQLite FTS5 SNOMED CT ATC Drugs LOINC

🔍 heart attack

Myocardial infarction

SNOMED · 22298006

Acute coronary syndrome

SNOMED · 394659003

Cardiac arrest

SNOMED · 410429000

[ MODE: RESCUE_RADAR ]

⚠️ Victim Found (12m)

Haru N. • Blood: A+ • Crit: High

🟡 Victim Found (28m)

Kenzo T. • Diabetes T1

Emergency Response

P2P Triage Mesh

JemmaPass transforms a collection of smartphones into a localized medical mesh. Rescuers can perform collaborative triage (SALT protocol), tagging victims as Green, Yellow, Red, or Black. These tags propagate across all Nearby devices instantly, ensuring no one is triaged twice.

✓ SALT Triage tagging & propagation
✓ Victim profile reassembly (multi-chunk)
✓ Rescuer capability discovery

🆚 Comparison

Jemma vs. the alternatives

No other solution combines an on-device AI companion with a portable clinical standard and full air-gap capability.

Feature	JemmaPass	Apple Health	Paper Records
Offline / Air-Gap	✓ Full	✗ iCloud only	✓ Physical
FHIR R4 compliant	✓	✓ Partial	✗
QR readable by paramedics	✓	✗	✗
Multilingual AI summaries	✓ 6+	✗	✗
On-device AI companion	✓ Gemma 4	✗	✗
Voice & Camera input	✓ ML Kit	✗	✗
Nearby SOS / Mesh	✓ On-device	✗	✗
DDI / DFI safety check	✓ On-device	✗	✗
Free forever	✓	✓	✓
No account needed	✓	✗ Apple ID	✓

Honest feature audit — Demo vs. Roadmap

Gemma 4 On-Device Chat

68 Gemma @Tool Functions

Med-Scan AI Pipeline

Voice Input (STT)

Text-to-Speech (TTS)

Multilingual UI (FR / JA)

3.36 GB Clinical KB — DDInter 2.0

DDI + Allergy + Drug-Disease Cross-check

Red Alert UI Trigger

3 Active Pillars

14 Passive Pillars

FHIR R4 IPS Bundle Export

Triple-Layer QR Code

Human-readable IPS Text

Live QR Scan (Camera)

SOS Broadcast (Victim Mode)

Rescue Radar (Rescuer Mode)

SALT Triage Protocol

BLE Fallback (Victim Discovery)

Emergency Home Screen Widget

Mesh Relay & Multi-hop

100% Offline & Air-Gap

Clinical Knowledge Base

Triple-Layer QR

Multilingual AI Summaries

Privacy by Design

Nearby · SOS Mesh

Built on HL7 IPS —the global standard

The guardian that never sleeps

Ask Jemma anything —she finds the right answer

P2P Triage Mesh

Jemma vs. the alternatives

Built on HL7 IPS —
the global standard

The guardian that
never sleeps

Ask Jemma anything —
she finds the right answer