跪拜 Guibai
← All articles
Vue.js · GitHub · Frontend

MedDoc AI: One Workbench for Clinical Writing, Collaboration, and Evidence-Based AI

By 徐小夕 ·
Read original on juejin.cn ↗ Google Translate ↗ Alt translation

Medical software demands trust and traceability more than speed: this design shows how to make AI genuinely usable in clinical settings by tying every answer to a cited source and enforcing permissions at the data layer, a pattern that applies beyond China's hospital IT environment. For Western developers building regulated or high-stakes AI tools, the lesson is that evidence-grounded, permission-aware workflows beat flashy automations.

Summary

Clinical staff waste hours toggling between hospital systems to write a single progress note, and existing AI tools often generate content without a traceable evidence chain, which is a non-starter for medical records that carry legal weight. MedDoc AI, built on the JitWord collaborative document engine, consolidates writing, multi-person editing, quality control, and knowledge Q&A into one workspace, with permissions enforced at the backend and every action logged.

The workbench includes a doctor station that aggregates to-dos and deadline warnings, a collaborative editor that preserves full version history, and a patient 360 profile that timelines all authorized data. Quality control rules run live during writing, flagging missing required fields, time limit issues, and format problems before archiving. Its AI answer feature retrieves evidence within the knowledge base and authorized records first, cites sources on every conclusion, and explicitly says when information is insufficient rather than fabricating.

For imaging, DICOM files support 3D volume rendering and three-plane linkage, while lab results are displayed as trend and scatter charts with abnormal values highlighted. The system supports private on-premise deployment, keeping all data inside the institution, and every feature is designed around role-based access and full audit trails. The goal is not to replace physicians but to offload mechanical documentation work, leaving judgment to humans.

Takeaways
— MedDoc AI is built on JitWord, a collaborative AI document foundation, and integrates clinical writing, collaboration, quality control, and knowledge Q&A into one workbench.
— Real-time quality control checks required fields, writing deadlines, and format rules during the writing process, allowing fixes before archiving.
— The AI answer feature retrieves evidence from the knowledge base and authorized records, cites source documents and sections for every conclusion, and declares when evidence is insufficient.
— Role-based access control is enforced at the backend, not just in the UI; all document modifications, workflow actions, and data access are logged and auditable.
— DICOM imaging supports 3D volume rendering and simultaneous axial, coronal, and sagittal viewing; lab data is visualized with trend and scatter charts, with abnormal values highlighted.
— Private on-premise deployment is supported, keeping all medical data within the institution's own environment.
— The complete workflow spans document creation, online collaboration, quality control review, and traceable archiving, all within a single platform.
— Structured templates for admission notes, progress notes, and consultation opinions standardize writing and speed up onboarding.
Conclusions

The 'conservative' approach—AI only assists with retrieval and organization, never fully automates writing—is a deliberate trust-building strategy that many production AI tools overlook.

The biggest time sink for clinicians isn't writing itself but the context-switching between disparate systems; aggregating views is worth more than clever AI features.

Moving quality control from post-hoc sampling to real-time inline checks fundamentally reduces rework, a lesson that applies to any compliance-heavy workflow.

Backend permission enforcement, rather than frontend hiding, is the only acceptable way to handle sensitive data; this is a benchmark for medical or any regulated software.

Concepts & terms
JitWord
A collaborative AI document foundation that provides online editing, version tracking, and real-time co-authoring capabilities, forming the base layer for MedDoc AI's document features.
DICOM
Digital Imaging and Communications in Medicine; the international standard for medical imaging data and communication, used to store and transmit CT, MRI, and other scan images.
Evidence-based AI (in medicine)
An AI approach where every generated claim or answer is backed by retrievable source documents or data, so users can verify the information; used here to avoid medical hallucinations and compliance risk.
Source: juejin.cn ↗ Google Translate ↗ Backup ↗