For Licensed Clinicians · Android

AI-assisted differential diagnosis at the point of care

Second Opinion connects doctors to a medically-oriented AI model during consultations — surfacing structured differentials, evidence citations, and red-flag alerts in real time.

Explore Features

Clinical decision support only · Not a medical device · For licensed clinicians

Second Opinion Clinical Decision Support Cloud-primary Aafiya AI workflow AI ONLINE INPUT MODES Text Query Type a clinical question Voice Query Speak a one-shot question Consultation Mode Passive background listening Image Analysis Photo or camera capture Text Query Describe the clinical presentation 45F, pleuritic chest pain × 3 days. HR 108, SpO₂ 93% on air. Long-haul flight 5 days ago. No fever or cough. 84 / 500 characters Ask Aafiya AI TIPS Include: age, sex, symptoms, duration, vitals, and relevant history. Avoid patient-identifiable details. Voice Query Speak your clinical question Listening... Tap circle to stop recording Consultation LIVE TRANSCRIPT MEDGEMMA SUGGESTION Top differential (updating...) Pulmonary embolism Pleurisy / viral pleuritis Pneumonia with pleuritic View full →
Input Modes

Four ways to engage the AI

Each mode is optimised for a different point in the clinical workflow — from the waiting room to the bedside.

Text Query

Type a clinical question — presentation, vitals, history snippet — and receive a fully structured differential diagnosis response within seconds.

Voice Query

Speak a one-shot diagnostic question hands-free. The app transcribes on-device and fires the query immediately — no typing required.

Consultation Mode

Runs passively in the background during a live consult. The app listens, builds a transcript, and surfaces differential suggestions without interrupting the conversation.

AI

Image Analysis

Photograph skin lesions, wounds, or visible anatomy from gallery or camera. The multimodal AI analyses the image and returns a differential with evidence reasoning.

Real-world Scenarios

Designed for the real ward

Second Opinion fits into existing clinical workflows — not alongside them.

Voice Query
Emergency Department

During a fast-moving ED shift with simultaneous resuscitations, query differential diagnoses hands-free between procedures — without breaking workflow or touching a screen.

Hands-free · Responds in seconds
Text Query
Rural General Practice

In a remote clinic without immediate specialist backup, get ranked differentials and red-flag alerts for complex presentations — when specialist review is hours or days away.

Works anywhere with signal
Consultation Mode
Teaching Ward Round

During a teaching presentation, let the app listen passively and build its own differential. Compare the AI's output with the team's assessment — a live, low-effort teaching tool.

Passive listening · Sparks discussion
Live Preview

See a real AI response

Simulated output for a typical high-stakes presentation — suspected pulmonary embolism.

Text Query

45F, acute onset pleuritic chest pain and progressive dyspnoea × 3 days. HR 108, RR 22, SpO₂ 93% on air. Recent long-haul flight 5 days ago. No fever or productive cough.

Sent to Aafiya AI
Aafiya AI · Second Opinion
01
Differential Diagnosis
  1. Pulmonary embolism high probability — Wells 6
  2. Pleurisy / viral pleuritis
  3. Pneumonia with pleuritic component
  4. Spontaneous pneumothorax
02
Supporting Evidence

Tachycardia (HR 108), hypoxia (SpO₂ 93%), pleuritic pain, and recent prolonged immobilisation meet Wells criteria ≥4. PE is clinically the most likely diagnosis.

03
Suggested Investigations

CTPA (first line); D-dimer only if low pre-test probability; ECG, ABG; FBC, CRP; troponin for risk stratification if PE confirmed.

04
⚠ Red Flags

SpO₂ 93% with tachycardia — haemodynamic instability risk. Escalate immediately if BP drops or GCS changes. Massive PE must be excluded urgently.

05
Disposition

Admit urgently. Initiate empiric anticoagulation (LMWH) if CTPA delayed and no contraindications. Senior review within 30 min. Consider HDU admission.

Illustrative output only. All AI suggestions require independent clinical verification by the treating clinician.

Structured AI Output

Every response has the same five sections

Consistent structure means you always know where to look — even in a high-pressure consult.

Section 01

Differential Diagnosis

Ranked list of likely diagnoses, ordered by probability given the clinical context.

Section 02

Supporting Evidence

Key features from the history, exam, or image that support each differential item.

Section 03

Suggested Investigations

Recommended bloods, imaging, or bedside tests to narrow the differential further.

Section 04

Red Flags

High-acuity findings or patterns that warrant urgent escalation or senior review.

Section 05

Disposition Guidance

Whether to admit, refer, or manage outpatient — with explicit uncertainty where appropriate.

Architecture

Where your data goes — and doesn't go

A round-trip to a single endpoint. No intermediaries, no analytics, no storage.

Your Device Query entered HTTPS Request Inference Engine Medical AI inference Structured Response Your Device 5-section output Data is not stored, logged, or shared at any point in this flow.
TLS-encrypted in transit No data stored server-side No third-party analytics Single configured endpoint only
How It Works

From question to structured output in seconds

The app routes your query to a cloud-hosted Aafiya AI inference endpoint — no local model required for the primary workflow.

Open the app

Accept the one-time clinical disclaimer. The AI service initialises in the background and signals when it is ready.

Choose a mode

Select Text, Voice, Consultation, or Image Analysis from the home screen dashboard.

Enter your query

Type, speak, or photograph. The app handles transcription and image encoding before dispatch.

Query dispatched

A structured clinical prompt wraps your input and is sent to the Aafiya AI inference engine endpoint via encrypted HTTPS.

Receive the differential

The five-section response streams back — differential, evidence, investigations, red flags, and disposition.

Apply clinical judgement

Review the output as a second opinion. Final diagnostic and treatment decisions remain entirely with the treating clinician.

Privacy & Security

Built with clinical data sensitivity in mind

No patient identifiers are stored. No backups. No third-party analytics.

Encrypted Storage

Credentials are held in Jetpack Security EncryptedSharedPreferences with AES256-GCM keys — never written to plain storage.

In-memory Transcripts Only

Consultation transcripts exist only in device RAM and are discarded the moment the session ends — nothing is written to disk.

No Device Backups

android:allowBackup="false" prevents any app data from appearing in Android or Google backups.

HTTPS-only Network

Network Security Config restricts all traffic to HTTPS. Inference only reaches the configured Aafiya AI endpoint.

No Patient Identifiers

The app does not collect, store, or transmit any patient-identifiable information. Clinicians are responsible for de-identifying queries.

One-time Disclaimer

Every install shows a clinical disclaimer on first launch. Acceptance is stored locally in DataStore — re-accepted on fresh install.

FAQ

Common questions

Things clinicians and IT departments typically ask before deploying a new clinical tool.

Aafiya AI is a medically-oriented language model but it is not validated for clinical use and has not undergone regulatory review. All outputs are general-model suggestions informed by a structured medical system prompt. They must be critically evaluated by the treating clinician and must not be acted on without independent verification. Think of it as a well-read colleague who reads quickly — not a diagnostic oracle.
No. Consultation transcripts are held only in device RAM for the duration of the session and cleared immediately when it ends — nothing is written to disk. The app has android:allowBackup="false" set, preventing Google backups. The only persistent data is your HuggingFace access token (for the optional on-device model download), stored in AES256 EncryptedSharedPreferences and never transmitted beyond huggingface.co.
The primary cloud-inference path requires an internet connection to reach the configured Aafiya AI endpoint. An optional on-device model (Gemma 3n, ~3 GB download) can be set up from the home screen — once downloaded, it enables offline differential support without any network call. Note: the on-device model is a general-purpose model without medical fine-tuning; clinical quality depends on the system prompt.
The home screen displays a real-time AI status indicator. If the Aafiya AI endpoint is unreachable, the app signals this clearly ("Config Needed" or "Attention Required") and queries will not process. If the optional on-device Gemma 3n model has been downloaded, the app can fall back to local inference automatically. Otherwise, the clinician is informed and no silent failures occur.
Second Opinion is designed for any licensed clinician who needs rapid differential support at the point of care. It is particularly well-suited to emergency medicine, general practice, internal medicine, and rural or resource-limited settings. Dermatology workflows benefit from image analysis. Teaching environments benefit from consultation mode during ward rounds.
No. Second Opinion is a clinical decision-support tool only and has not been approved, cleared, or certified by any regulatory authority — including the FDA, CE, SAHPRA, or equivalent. It is not intended to replace clinical judgement and must not be used as the sole basis for a diagnostic or treatment decision. Final clinical responsibility rests entirely with the treating clinician.
Get in Touch

Contact the SecondOpinion Team

Have questions about deployment, integration, or clinical use cases? We're here to help.

Requirements

What you need to run Second Opinion

Android 8.0+ (API 26) Physical device (on-device AI path) Internet (cloud inference) Aafiya AI endpoint configured ~3 GB storage (on-device, optional) 4 GB+ RAM recommended
Ready to put AI in your pocket?

Cloud-primary, no model download required. Get structured differentials in seconds on any Android 8.0+ device.

For licensed clinicians only · Clinical decision support · Not a medical device

Clinical Decision Support Only

Second Opinion is a clinical decision-support tool, not a medical device. It has not been approved, cleared, or certified for diagnostic use by any regulatory authority. All AI-generated suggestions are produced by a general-purpose language model and must be critically evaluated by the treating clinician before any clinical decision is made. Final diagnostic and treatment responsibility rests entirely with the treating physician. Do not enter patient-identifiable information into the app.