HF Readmit Agent

HF Readmit Agent is a self-reported research prototype built for evaluating heart-failure follow-up triage using synthetic data. The project is described as an evidence-first, human-in-the-loop workstation that allows reviewers to inspect, challenge, and measure escalations from deterministic rules or optional LLMs. It is not a medical device or clinical decision-support system.

The author states the system supports a reviewer-specific case queue, daily symptom timelines, traceable triage tiers (L0–L3), and analytics comparing rule-based and LLM outputs. The prototype uses React + Go + PostgreSQL, Docker Compose for local deployment, and includes an optional GPT-5.6 second reader constrained by JSON schema and citation verification.

Key claims include: the system is inspectable, challengeable, and measurable; it preserves human judgment; and it avoids clinical truth claims or untested assertions. It is strictly a research tool with no real patient data or regulatory use.

The description states that HF Readmit Agent is an evidence-first, human-in-the-loop workstation for evaluating post-discharge heart-failure follow-up triage on a cohort of 18 synthetic cases.

It allows reviewers to:

• Open a reviewer-specific case queue and inspect a compact patient snapshot.
• Review a daily timeline of symptoms, vitals, adherence, and trends.
• See the deterministic L0–L3 triage tier, named rule, source day, and source value behind each escalation.
• Record an independent final tier and explain agreement, modification, or disagreement.
• Explore analytics and Safety Lab results that compare deterministic rules, an optional LLM second reader, and human reviewers.

It is described as a research prototype, not a medical device, diagnostic tool, or clinical decision-support system. It uses synthetic data only and does not use real patient data.

Inference The product appears to be a local, reproducible research environment for evaluating triage logic and LLM behavior in a controlled setting.

The author states that HF Readmit Agent was inspired by the question: “how can AI support evaluation without hiding the evidence, replacing clinical judgment, or making untested claims?”

It positions itself as:

• Evidence-first.
• Human-in-the-loop.
• Inspectable, challengeable, and measurable.

The project is framed as a research workflow, not a product for clinical deployment. It explicitly states it does not use real patient data and is not a medical device or diagnostic tool.

Inference The positioning reflects an intent to explore AI-assisted triage in a controlled, transparent way — not to replace human judgment but to support it with traceable evidence.

The description states that HF Readmit Agent is built for reviewers, specifically those evaluating heart-failure follow-up triage. These reviewers are described as:

• Clinicians or researchers working in a triage context.
• Users of a reviewer-specific case queue.
• Individuals who can inspect, challenge, and measure escalations.

It is not stated whether the target customer is a healthcare organization, research team, or individual developer.

Inference The ICP is likely healthcare professionals or researchers involved in evaluating triage systems — especially those interested in AI-assisted decision-making with transparency and traceability.

The description states that HF Readmit Agent is a research prototype, not a commercial product. It does not mention any pricing, licensing, or monetization strategy.

It is explicitly stated that the system:

• Does not use real patient data.
• Is not a medical device or clinical decision-support system.
• Is strictly for research and evaluation purposes.

Inference No business model or pricing evidence is provided. The project is self-described as a prototype with no commercial intent.

The author states that HF Readmit Agent was built as a full-stack prototype using:

• Frontend: React + Vite + TypeScript.
• Backend: Go for REST API, deterministic rules engine, reviewer queues, server-side review timing, exports, analytics, and LLM safety controls.
• Database: PostgreSQL.
• Deployment: Docker Compose for local stack reproduction.
• AI Integration: Optional GPT-5.6 second reader with structured output, citation checks, and server-side verification.

It also includes:

• A deterministic rules engine (v1.0) that evaluates synthetic follow-up days and returns L0–L3 results with named, evidence-producing rules.
• A Safety Lab for comparing deterministic rules, optional LLMs, and human reviewers.
• Bilingual interface, theme settings, SVG icons, light/dark mode support.

Inference The technical stack is self-contained, reproducible, and designed for local deployment. It includes safety controls around LLM use.

The description states that HF Readmit Agent is a research prototype, not a product in production or with live users.

It does not provide evidence of:

• Revenue.
• Customers.
• Adoption.
• Product usage metrics.
• Real-world deployment.

It is described as a self-contained, runnable system for immediate testing without API keys or accounts.

Inference There is no traction or maturity signal beyond the prototype’s own development and demonstration. No external validation or user feedback is reported.

The description does not mention any direct competitors or competitive landscape.

It is framed as a research tool, not a commercial product, so there is no stated market position or competitive advantage in a marketplace context.

Inference The project operates in a niche research space and does not appear to be competing with existing triage or AI-assisted clinical tools. No competitive signals are evident.

• No real-world use: The system is strictly a research prototype, not a product for clinical deployment.
• No external validation: There is no evidence of usability testing with clinicians beyond the prototype’s own design.
• Unverified LLM safety: While the author claims server-side controls, there is no independent verification or audit of these mechanisms.
• Synthetic-only data: The system uses synthetic data only; it does not reflect real-world complexity or variability.
• No commercial intent: No evidence of a path to monetization or product-market fit.

Inference The project’s value lies in its research utility, but it is not a viable commercial or clinical product without further development and validation.

1. What are the specific use cases for which this prototype was designed?
2. Has it been tested with actual clinicians or researchers?
3. How does the system handle edge cases or unexpected inputs in synthetic data?
4. Are there any plans to move beyond synthetic-only data?
5. What is the intended path from research to clinical deployment, if any?
6. How are the deterministic rules validated or developed?
7. What are the limitations of the current LLM integration and how are they mitigated?