Skip to content

Huantuk

Huantuk is : 1.

Code Video Cerebral Valley

Core trick

Extracts structured case objects from redacted reports, then matches unresolved cases against guideline and trial literature for repurposed drug candidates.

Tags

healthscience

Description

Huantuk is : 1. Diagnosis synthesizer : input health reports and clinical history, after first understanding, answer further clinical history and input further health reports (as recommended and other new health reports), output diagnosis and current clinical practice guideline (CPG) management 2. Drug discovery : input huantuk if unresolved after CPG management, read through all personalized health datas, research through medical publications, clinical trials and related evidence based articles, output repurposed drug, off-label indicated medication and applicable clinical trials
RC24bc/huantuk★ 00 forksTypeScriptNOASSERTIONCreated Apr 21, 2026 · last push Apr 26, 2026

README excerpt

# Huantuk · Patient Atlas

> *"Illnesses which in the past caused exorcism — or what traditional Chinese would say, huantuk (撞鬼) — that are difficult to diagnose but Claude can solve."*

**Huantuk** is a clinical decision-support tool for autoimmune cases that don't fit the textbook. Built on **Claude Opus 4.7** for the Cerebral Valley × Anthropic hackathon. Live at [huantuk.vercel.app](https://huantuk.vercel.app).

## Two functions, one workflow

### 1 · Diagnostic Synthesizer

For patients with months of specialist visits and stacks of reports, but **no diagnosis**. Feed the folder — lab panels, imaging reports, clinic letters, clinical photos — from any combination of hospitals and Huantuk:

1. **Extracts + redacts** PII from each document, returns a structured case object.
2. **Synthesises** a 3-sentence narrative summary in neutral clinical register.
3. **Ranks the top 5 differentials** with posterior probability, supporting + contradicting findings, and citations to the source paper for every claim.
4. **Scores classification criteria** met / unmet / unknown against seven published sets — Yamaguchi AOSD, 2017 EULAR/ACR IIM, 2019 ACR/EULAR SLE, 2010 ACR/EULAR RA, 2022 ACR/EULAR AAV, 2019 ACR/EULAR IgG4-RD, 2016 ACR/EULAR Sjögren's.
5. **Screens for autoimmune mimics** — lymphoma, chronic infection, drug reaction, VEXAS, IgG4-RD masquerade — that get missed when each specialist sees only their own slice.
6. **Ranks the next tests** by expected information gain over a 24-t

Team