🏆 GEMMA 4 CONTEST DEMO — All implemented features are fully functional. 15 IPS pillars are planned for production.
✦ The Silicon Shell Architecture

Intelligence that stays local

Discover how JemmaPass uses a Dual Edge Architecture (Gemma 4 LiteRT-LM & Chrome Built-in AI) to create the world's most resilient clinical data engine — 100% on-device, zero network required.

👁️
Perception
ML Kit · Vision
🧠
Reasoning
Gemma 4 · LiteRT
🛡️
Clinical Shield
3.36 GB SQLite · FTS5
⚒️
Clinical Forge
IPS · FHIR R4
📶
Transfer
Nearby · Triple QR
1
Phase 01 · Perception
Multimodal Onboarding

Forget long forms. Talk to Jemma or point your camera at your medication boxes. JemmaPass uses Google ML Kit Document Scanner (which requires zero camera permissions!) to scan labels and voice intent parsing to understand your symptoms and history instantly.

🎙️
Audio Scribe + OCR Vision → Local Buffer
STYLE: Chibi 3D illustration, dark navy bg, teal laser scan line. | CHARACTER: Jemma chibi turtle. | SCENE: Jemma holding a tiny glowing lens over a medication bottle, data stream flowing into her shell. Expression: high-tech precision. | --ar 16:9 --v 6.0
2
Phase 02 · Reasoning
Gemma 4 Local Inference

The data is processed by Gemma 4 running locally via LiteRT. The AI companion maps your inputs to professional clinical standards (SNOMED CT, RxNorm) autonomously, without ever calling a cloud server.

🧠
Gemma 4 @Tools · Function Calling · On-Device Reasoning
STYLE: Chibi 3D illustration, dark navy bg, indigo node network. | CHARACTER: Jemma chibi turtle. | SCENE: Jemma floating in a network of glowing medical nodes, connecting dots with her paws. Expression: enlightened. | --ar 16:9 --v 6.0
3
Phase 03 · Shielding
Clinical Shield Safety Check

JemmaPass queries its 3.36 GB Knowledge Base to perform Drug-Drug (DDI) and Drug-Food (DFI) interaction checks. She warns you instantly if a new medication conflicts with your existing profile or allergies, drawing from a comprehensive clinical universe.

🛡️
SQLite FTS5 · Trigram Search · SNOMED CT Cross-Ref
STYLE: Ultra-clean chibi 3D illustration, tight circular 1:1 composition, protective amber/indigo glow, dark navy background. | CHARACTER: Jemma chibi turtle, mint-green skin, large teal eyes, dark teal shell with LED cross. | SCENE: Jemma holds up a translucent indigo digital shield. On the other side of the shield, a red medication bottle icon bounces off harmlessly. A glowing 'SAFE' checkmark appears in teal. Expression: protective and vigilant. | NEGATIVE: Realistic, human, text, watermark. | --ar 1:1 --v 6.0 --stylize 180
4
Phase 04 · Forging
IPS FHIR Compression

The Clinical Forge structures your data into an International Patient Summary (IPS). We use advanced clinical pruning to strip heavy metadata, reducing the typical bundle from 1009 bytes raw → 500 bytes wire (~50%), while keeping it 100% FHIR-interoperable.

⚒️
HL7 FHIR R4 · deflate-raw (RFC 1951) · Clinical Pruning ~50%
STYLE: Ultra-clean chibi 3D illustration, tight circular 1:1 composition, emerald furnace glow, high-tech sparks, dark navy background. | CHARACTER: Jemma chibi turtle. | SCENE: Jemma stands at a glowing digital anvil. She strikes a heavy FHIR data block with a glowing hammer, causing it to compress into a dense, sparkling teal QR code. Tiny 'raw data' bits fly off (pruning). Expression: master artisan focus. | NEGATIVE: Realistic, human, text, watermark. | --ar 1:1 --v 6.0 --stylize 200
5
Phase 05 · Transfer
Nearby Mesh & Triple-Layer QR

Your profile is ready for the world. Share it via Nearby SOS Broadcast for instant p2p transfer, or display the Triple-Layer QR (Binary, Human, Fallback). The mesh allows rescuers to find you even without cellular signal.

📶
Nearby Connections · P2P Cluster · Mesh Relay
STYLE: Ultra-clean chibi 3D illustration, cinematic depth of field, dark navy background (#050911), vibrant emerald and indigo glow accents. | CHARACTER: Jemma chibi turtle, mint-green skin (#CCFFCC), large sparkling teal eyes, dark teal shell (#008080) with circuit LED cross on back, small glowing QR code on front chest plastron emitting teal light. | SCENE: Jemma leans forward in a dynamic heroic pose, tapping her Medical Smartphone against a second device. At the contact point, a powerful burst of teal light and electric NFC pulse rings radiate outward. Translucent FHIR data packets visibly travel between the devices through the light. Expression: triumphant and joyful. | NEGATIVE: Realistic photography, human face, text overlay, watermark, blurry, low quality, flat 2D cartoon, adult content, realistic turtle. | --ar 16:9 --v 6.0 --stylize 250
🧠
GEMMA 4 · ACTIVE
"I've detected a conflict between Warfarin and your recent Aspirin scan. Clinical Shield suggests consulting a doctor immediately."
View DDI Report

Real-time on-device reasoning demo

❓ Technology FAQ

Inside the Shell

How can Gemma 4 run on a phone? ▼
JemmaPass uses a dual Edge AI architecture: Native Android Wrapper with LiteRT-LM (Gemma 4 E4B at 3.4 GB, or E2B at 2.4 GB as .litertlm files hosted directly on jemmapass.com), and an automatic fallback to Chrome Built-in AI (Nano) on the web.
What is "Clinical Pruning"? ▼
Raw FHIR JSON contains 70% "noise" (metadata, URLs, formatting). JemmaPass\'s Clinical Forge strips this away while keeping the core clinical codes. This allows a complete 18-pillar IPS profile to fit into a single high-density QR code.
Demo note: 3 active pillars, 15 reference stubs.
How does the Rescue Radar work? ▼
Jemma uses Google Nearby Connections in P2P_CLUSTER mode. Devices broadcast their status in rotating 131-byte chunks. Rescuers reassemble these chunks into full clinical profiles and display them on a radar using local GPS data (all offline).
Why 3.36 GB for the database? ▼
Unlike basic apps, JemmaPass embeds a **full clinical universe**. This 3.36 GB "Clinical Shield" contains the complete SNOMED CT and RxNorm ontologies, optimized with SQLite FTS5 for sub-100ms searches. It ensures clinical-grade safety with zero cloud dependencies.