Clinical Referral Copilot

The company appears to be a solo project (1 person) submitted to the OpenAI 2026 hackathon, titled Clinical Referral Copilot. The author describes it as an AI-powered tool for structuring and triaging specialist referrals in gastroenterology, with a strong emphasis on transparency, clinician control, and traceability. It is built using GPT-5.6 via the OpenAI Responses API, Next.js, TypeScript, and React.

What changed: The project is described as a demonstration of a safer pattern for healthcare AI — one that structures and explains decisions while keeping clinicians in the loop. It avoids persuasive or autonomous behavior by making uncertainty, provenance, and refusal boundaries explicit.

Single most important open question: Is there any evidence of real-world usage, clinician feedback, or integration into actual clinical workflows? The description is entirely self-reported and lacks any data on adoption, revenue, or customer engagement beyond the synthetic demo.

The description states that Clinical Referral Copilot:

• Turns a synthetic gastroenterology referral into:
  • Structured clinical facts
  • Recommended urgency and timeframe
  • Evidence linked to exact phrases in the referral
  • Transparent synthetic rule labels
  • Missing information and uncertainty
  • Suggested pathway
  • Explicit clinician approve/edit/reject decision

• Includes a small synthetic evaluation dashboard, safety case, identifier blocking, and narrative-free session audit trail.

• Is built as a clean-room Next.js and TypeScript application using GPT-5.6 via the OpenAI Responses API.

• Is explicitly synthetic-only and takes no autonomous action.

Inference: The product is a frontend UI for reviewing structured outputs from an AI model, with a focus on transparency and clinician involvement. It does not appear to be a production-ready system but rather a prototype or demo.

The author states:

• “Clinical Referral Copilot explores a safer pattern for healthcare AI: AI structures and explains; a clinician decides.”

• The project avoids a persuasive but unsafe “AI doctor” demo by making uncertainty, provenance, refusal boundaries, and clinician authority visible.

• It is described as a demonstration of how trust in healthcare AI can be built through inspectability, not confidence scores.

Inference: The positioning is centered on safety, transparency, and clinician control, rather than automation or efficiency. It positions itself as an alternative to traditional AI decision-making tools in healthcare.

The description states:

• The tool is designed for clinicians reviewing specialist referrals (specifically gastroenterology).

• It is built for rapid clinical review, with a responsive, accessible interface.

• The user is expected to approve, edit or reject decisions made by the AI.

Inference: The ICP appears to be clinicians in specialty referral settings, particularly those dealing with high-volume, repetitive triage tasks. No evidence of other personas or use cases is provided.

The description does not state:

• Any pricing model
• Revenue streams
• Customer acquisition plans
• Monetization strategy

Not evidenced: There is no indication of a business model or pricing structure beyond the project being a hackathon submission.

The author states:

• Built with: codex, gpt-5.6, json-schema, next.js, openai-responses-api, react, typescript
• Uses GPT-5.6 via OpenAI Responses API with medium reasoning effort, store: false, and strict JSON Schema output
• Server-side route calls the API
• UI renders predictable structured fields
• Allows cross-highlighting of model reasons against source phrases
• Includes deterministic demonstration engine
• Blocks common Singapore identifiers before processing
• App is synthetic-only and takes no autonomous action

Inference: The technical stack suggests a frontend-heavy, AI-assisted workflow tool, with strong emphasis on structured output control and traceability. It is not a scalable or production-grade system but a prototype.

The description states:

• This is a hackathon submission
• The app is synthetic-only
• No real-world usage, customers, or adoption data are mentioned
• The next step is to evaluate representative synthetic cases and undergo privacy, cybersecurity, clinical-safety, and governance review

Not evidenced: There is no evidence of traction, revenue, customer engagement, or product maturity beyond a prototype.

The description does not mention:

• Competitors
• Existing tools in the referral triage space
• Market size or competitive landscape

Not evidenced: No competitive context is provided. The project appears to be self-contained and not positioned against existing solutions.

• No real-world usage or feedback: The product is described as synthetic-only, with no evidence of actual clinical use.
• Solo team: Only one person is listed on the project; no indication of a larger team or support structure.
• Hackathon prototype: The submission is from a hackathon, suggesting it is not yet mature for production.
• No monetization strategy: No business model or revenue plan is evident.
• Unverified claims: All claims are self-reported and unverified.

Inference: The project is in an early stage of development and lacks any commercial or clinical traction. It may be a proof-of-concept, not a viable product for investment or partnership.

1. What specific clinical workflows does this tool aim to support beyond gastroenterology?
2. How is the synthetic data being generated, and what are the limitations of that approach?
3. Has there been any feedback from actual clinicians on the interface or decision-making process?
4. Are there plans to move beyond a demo into a pilot or prototype with real users?
5. What would be the next steps in terms of clinical validation, regulatory compliance, or product development?