RESEARCH DEMO · NOT FOR DIAGNOSIS OR TREATMENT · HAI-DEF TERMS APPLY

A chronic pain management app designed for fibromyalgia

I adapted the design around the daily realities of fibromyalgia. HelloFibro brings supportive conversation, medication routines, appointment preparation, and symptom support into one calm interface.

Run live demo

Checking whether the live demo is available…

01 · THE PROBLEM

Designed around difficult days

Fibromyalgia is unpredictable and cognitively exhausting. Symptoms change from day to day, and brain fog can make even simple tasks feel difficult.

I focused the design on clarity, low effort, and useful structure for people managing pain, fatigue, medication routines, and medical appointments.

CHALLENGESymptoms change from day to day
DESIGN RESPONSEClear paths back into the experience
CHALLENGEPain and fatigue reduce available energy
DESIGN RESPONSELow-effort actions and short routines
CHALLENGEBrain fog makes decisions harder
DESIGN RESPONSEUseful structure with fewer choices

03 · THE SOLUTION

Practical support in one calm workspace

HelloFibro adapts familiar chronic pain management tools to fibromyalgia. The interface keeps daily tasks short, readable, and easy to return to when energy or concentration is limited.

01Daily workspaceChat · routines · preparation
02Safety and accessGuardrails · languages · privacy
03Context and inferenceModels · files · medical context
CHAT

Supportive conversation

Warm, structured responses help users work through symptoms, routines, and questions at their own pace.

ROUTINE

Medication tracking

Add medications, mark doses taken, and see adherence at a glance with low-friction interactions.

PREP

Appointments

Track upcoming visits and prepare focused questions for a clinician without leaving the flow.

GUARDRAILS

Clear safety boundaries

Urgent situations, model disclosure, and clinical limits remain visible throughout the experience.

ACCESS

Multilingual interface

The hosted UI offers complete interface translations for all 21 selectable locales. English is MedGemma's primary evaluated language; all other model-language paths remain experimental.

View all 21 supported UI languages
PRIVACY

In-memory prototype state

Conversation, medication, and appointment data remain in memory, with verified user isolation required before public launch.

INFERENCE

Medical model pathway

The hosted demo runs multimodal MedGemma 1.5 4B on Hugging Face ZeroGPU. Full multimodal MedGemma 27B inference requires a separate self-hosted/local GPU deployment.

FILES

File analysis

Users can discuss supported documents, spreadsheets, text files, and images in the same conversation, with validated uploads and extracted context.

MEDICAL CONTEXT

Fibromyalgia intelligence layer

Detected worsening factors, a structured knowledge service, and guideline checks add relevant context while preserving non-clinical boundaries.

ACCESS · LOCALIZATION

Supported UI languages

The live demo exposes 21 fully translated interface locales. Each matches the in-app language selector. English is MedGemma's primary evaluated language; all other model-language paths remain experimental.

Critical deterministic safety messages may fall back to English. A translated interface does not imply equivalent clinical validation of model output.

04 · LIVE PROOF

Open the protected demo

Hugging Face sign-in · MedGemma 1.5 4B · in-memory prototype data · professional medical guidance remains essential.

Research demonstration only. Do not use it for diagnosis, treatment, emergencies, automated healthcare decisions, or professional medical practice. The hosted version uses the smaller MedGemma 1.5 4B model. Full multimodal MedGemma 27B requires a separate self-hosted/local GPU deployment.

Run live demo

Checking whether the live demo is available…

05 · ARCHITECTURE

How the application handles safety

The public deployment places red-flag checks before inference and sanitizes the completed response before it reaches the interface. User-isolated in-memory state is also required before launch.

PYTHON 3.10+ · GRADIO 6 · PYDANTIC 2 · OPENROUTER / MEDGEMMA · DOCKER

06 · SAFETY & TRUST

Scope and safety boundaries

VISIBLE PRODUCT CONTROLS

Support with clear context

  • Prototype data stays in memory. Public launch also requires verified user isolation.
  • The application displays its active model and provider.
CLINICAL BOUNDARY

Decisions stay with people

  • Health decisions stay with the user and qualified medical professionals.
  • Emergency situations require local emergency services.
  • This prototype has no clinical validation or HIPAA certification.
DEMO USE TERMS

MedGemma access has explicit conditions

  • The demo requires acceptance of the HAI-DEF terms before model use.
  • Diagnosis, treatment, emergencies, automated healthcare decisions, and unlicensed medical practice are outside its permitted scope.
  • Read the HAI-DEF Terms and Prohibited Use Policy.

07 · ENGINEERING HIGHLIGHTS

Implementation details

UIGradio 6 messages, streaming, responsive custom styling
DATAPydantic models, i18n JSON, session-scoped state
SECURITYRate limiting, input sanitization, API-key masking
AGENTSFeature-flagged CARE orchestrator with a deterministic public safety path
DEPLOYMENTProtected ZeroGPU demo with MedGemma 1.5 4B; full MedGemma 27B available through separate local GPU deployment

08 · INFRASTRUCTURE LANDSCAPE

A patient-facing application shaped by healthcare engineering

HelloFibro reflects my work across health AI UX, agent systems, and careful use of language models. It complements FhirFlame, my healthcare data platform project.

Discuss a project