Challenge Overview
Presented by Care GP
Context & problem: Australian GPs face 4,500+ MBS item numbers with frequent rule/caveat updates. Manual look-ups slow consults, cause coding errors, and increase audit exposure—directly impacting revenue, admin load, and compliance.
Why now: Digitisation of clinics, increasing scrutiny on ethical billing, and rapid AI progress make it viable to deliver real-time decision support at the point of care.
Your task is to design and prototype a system that turns free-text or voice input from a consultation into:
• accurate top-N MBS item recommendations,
• automatic rule/ethics checks with clear justifications,
• a ready-to-submit claim payload plus autopopulated clinical documents (e.g., referral, care plan).
Challenge Overview
Presented by Care GP
Context & problem: Australian GPs face 4,500+ MBS item numbers with frequent rule/caveat updates. Manual look-ups slow consults, cause coding errors, and increase audit exposure—directly impacting revenue, admin load, and compliance.
Why now: Digitisation of clinics, increasing scrutiny on ethical billing, and rapid AI progress make it viable to deliver real-time decision support at the point of care.
Your task is to design and prototype a system that turns free-text or voice input from a consultation into:
• accurate top-N MBS item recommendations,
• automatic rule/ethics checks with clear justifications,
• a ready-to-submit claim payload plus autopopulated clinical documents (e.g., referral, care plan).
Challenge Scope
Who this helps
Primary users: GPs, registrars, nurse practitioners.
Secondary users: Practice managers and billers (compliance & reporting).
Problem types
• Automation (reduce claim prep time).
• Decision support & compliance (mutual exclusivity, time thresholds, telehealth/after-hours rules).
• Documentation generation (referrals, care plans) tied to chosen codes.
• Operational insight (error/reject alerts, revenue).
Core components (examples, not prescriptions)
• Ingestion: streaming speech-to-text or free-text notes.
• Item recommender: retrieval + rules/LLM/ranking to surface likely MBS items.
• Ethical guardrails: rule engine (mutually exclusive codes, time-based thresholds, context checks) with human-readable rationales.
• Claim builder: constructs a stub claim payload (e.g., FHIR Claim ) plus item modifiers.
• Docs autopilot: generates referral/care-plan drafts linked to the same items.
• Compliance dashboard: errors, rejects, revenue by date/provider/item; exportable audit report.
Real-world constraints
• 7-day build window.
• No real PHI: use synthetic patients and synthetic transcripts.
• Prefer AU data residency assumptions in design; offline/demo-friendly mode acceptable.
• Integration may be simulated/stubbed (e.g., FHIR/HL7 mock endpoints).
Out of scope (for hackathon)
• Live submission to Medicare or real PMS production integration.
• Handling every MBS rule; focus on a clear, well-documented subset.
Who this helps
Primary users: GPs, registrars, nurse practitioners.
Secondary users: Practice managers and billers (compliance & reporting).
Problem types
• Automation (reduce claim prep time).
• Decision support & compliance (mutual exclusivity, time thresholds, telehealth/after-hours rules).
• Documentation generation (referrals, care plans) tied to chosen codes.
• Operational insight (error/reject alerts, revenue).
Core components (examples, not prescriptions)
• Ingestion: streaming speech-to-text or free-text notes.
• Item recommender: retrieval + rules/LLM/ranking to surface likely MBS items.
• Ethical guardrails: rule engine (mutually exclusive codes, time-based thresholds, context checks) with human-readable rationales.
• Claim builder: constructs a stub claim payload (e.g., FHIR Claim ) plus item modifiers.
• Docs autopilot: generates referral/care-plan drafts linked to the same items.
• Compliance dashboard: errors, rejects, revenue by date/provider/item; exportable audit report.
Real-world constraints
• 7-day build window.
• No real PHI: use synthetic patients and synthetic transcripts.
• Prefer AU data residency assumptions in design; offline/demo-friendly mode acceptable.
• Integration may be simulated/stubbed (e.g., FHIR/HL7 mock endpoints).
Out of scope (for hackathon)
• Live submission to Medicare or real PMS production integration.
• Handling every MBS rule; focus on a clear, well-documented subset.
Solution Requirements
Builder’s challenge: we evaluate working systems
working systems, not slide decks.
1. Working Prototype / Demo
A functional system, interface, agent, or workflow that:
• Accepts clinical text (typed or transcribed) and returns top-N item suggestions.
• Runs selected rule checks with explicit “why” (audit-ready notes).
• Produces a claim payload (stub) and generates at least one linked document (e.g., referral).
• Shows a small dashboard view with errors/reject simulations and revenue snapshots.
2. Live Demo or Demo Video (10–20 mins)
Show what it does, how it works, and why it matters (include a short, realistic consult flow).
• Technical Brief (≤5 pages)
• Technical details: architecture diagram; tools/models/APIs; rules covered; data flows; any PMS/FHIR stubs.
• Problem fit & value: how you reduce time/errors; who benefits; expected clinic-level impact.
• Limitations & next steps: known gaps, safety & privacy considerations, scaling plan.
Builder’s challenge: we evaluate working systems
working systems, not slide decks.
1. Working Prototype / Demo
A functional system, interface, agent, or workflow that:
• Accepts clinical text (typed or transcribed) and returns top-N item suggestions.
• Runs selected rule checks with explicit “why” (audit-ready notes).
• Produces a claim payload (stub) and generates at least one linked document (e.g., referral).
• Shows a small dashboard view with errors/reject simulations and revenue snapshots.
2. Live Demo or Demo Video (10–20 mins)
Show what it does, how it works, and why it matters (include a short, realistic consult flow).
• Technical Brief (≤5 pages)
• Technical details: architecture diagram; tools/models/APIs; rules covered; data flows; any PMS/FHIR stubs.
• Problem fit & value: how you reduce time/errors; who benefits; expected clinic-level impact.
• Limitations & next steps: known gaps, safety & privacy considerations, scaling plan.
Suggested Demo Flow
1. Signals to model: consultation length & content cues (telehealth vs。in-person; after-hours); mutually exclusive codes; time thresholds; required elements for care plans/assessments.
2. Rule handling options:
• Lightweight rules engine (YAML/JSON rules → human-readable rationales).
• Retrieval-augmented prompting for “why this code” explanations.
• Hybrid ranker: keyword/regex features + ML/LLM rerank.
3. Data & interfaces (mocked is fine):
• Synthetic MBS item catalogue (JSON with descriptors, fees, flags, timing).
• Synthetic transcripts (SOAP style) and patient/practitioner records.
• FHIR resources ( Patient , Practitioner , Encounter , Claim ) via a local stub.
4. UX ideas:
• Inline suggestions in a notes editor (“Accept/Swap/Explain”).
• Red/amber/green compliance banner with drill-down justifications.
• One-click “Create Claim” + “Generate Referral” linked to chosen items.
5. Helpful approaches:
• Few-shot prompts for rationale templates and document scaffolds.
• Guardrail tests (unit tests for rules) and a tiny eval set (10–20 consults) to track: top-3 code precision/recall on your chosen rule subset, rule check accuracy, median end-to-end latency.
1. Signals to model: consultation length & content cues (telehealth vs。in-person; after-hours); mutually exclusive codes; time thresholds; required elements for care plans/assessments.
2. Rule handling options:
• Lightweight rules engine (YAML/JSON rules → human-readable rationales).
• Retrieval-augmented prompting for “why this code” explanations.
• Hybrid ranker: keyword/regex features + ML/LLM rerank.
3. Data & interfaces (mocked is fine):
• Synthetic MBS item catalogue (JSON with descriptors, fees, flags, timing).
• Synthetic transcripts (SOAP style) and patient/practitioner records.
• FHIR resources ( Patient , Practitioner , Encounter , Claim ) via a local stub.
4. UX ideas:
• Inline suggestions in a notes editor (“Accept/Swap/Explain”).
• Red/amber/green compliance banner with drill-down justifications.
• One-click “Create Claim” + “Generate Referral” linked to chosen items.
5. Helpful approaches:
• Few-shot prompts for rationale templates and document scaffolds.
• Guardrail tests (unit tests for rules) and a tiny eval set (10–20 consults) to track: top-3 code precision/recall on your chosen rule subset, rule check accuracy, median end-to-end latency.
Prize
• Paid internship