@unisound-llm

unisound-surgery-review audits surgical procedure codes against structured medical records and rule libraries, emitting pass/fail/review-required verdicts with evidence-based reasoning. supports json, pdf, doc, xls, txt input formats and optional llm-assisted rule matching.

selects primary diagnosis and primary surgery from candidate lists by calling an internal medical llm, given admission records and candidate options. outputs medical coding assistance, not clinical advice.

audits surgical/procedural sufficiency against medical record evidence. takes structured case records and surgery lists, returns sufficiency verdicts based on guideline matching and document evidence extraction.

unisound-drg-grouping routes discharge summaries and drg candidate lists to an internal medical llm for single-choice classification, outputting results in structured json with optional metadata pass-through for evaluation workflows.

unisound-med-teach generates clinical teaching cases from question text or structured json input, routing them through an internal medical llm and returning formatted json output with answer, metadata, and compliance notes.

converts colloquial medical records to standardized clinical documentation via internal llm. supports multiple file formats, enforces data minimization and deidentification, outputs structured records across six canonical sections.

med-critical-disease-review evaluates critical illness insurance claims against 28 disease conditions by parsing structured hospital records and calling an internal assessment api, returning both raw json and natural language conclusions with supporting evidence.

unisound-chronic-injury routes insurance claim image reports through an internal medical LLM to classify injury age (old/new/none/unmentioned) against caller-defined categories and output format. no external dependencies, direct model invocation.

unisound-med-synonym matches medical terminology to canonical forms via internal llm. accepts question text or structured json input with configurable model parameters and multiple file input modes.

unisound-med-content-generation produces medical content across five task types (explain, popular, record-gen, summary, teach) via command-line task selection. inputs accept direct text, jsonl batches, json files, or stdin; output is json with status, task metadata, and model answer.

clinic-referral-guidance accepts patient case summaries and outputs structured referral decisions (urgency level, target department/hospital tier, pre-transfer measures, required documents) via internal medical llm, adhering to data minimization and immediate destruction principles.

unisound-med-clinical-diagnosis provides clinical case analysis, differential diagnosis, and primary care disease classification via task-switching. deployment is self-contained (scripts/run.py only) but relies on proprietary internal medical model access and appkey authentication.

unisound-paper-writing-assist expands research bullet points into imrad-structured academic paragraph drafts in chinese or english, with checklist reminders aligned to strobe/consort reporting standards. author retains responsibility for factual accuracy.

audits outpatient emr records for hypertension cases missing maximum blood pressure values by invoking a medical llm, returning defect status and reasoning. designed for chinese healthcare settings.

unisound-med-treat generates structured diagnostic and treatment recommendations from medical case summaries via internal llm, accepting question text or jsonl/json files with configurable model parameters and dry-run capability.

unisound-missing-tumor-name performs quality control on outpatient records by detecting missing tumor names via internal medical llm. accepts plain text or multiple document formats, outputs pass/fail with reasoning.

unisound-chief-complaint-diagnosis-inconsistent validates outpatient medical records by detecting inconsistencies between chief complaint and diagnosis location using an internal medical llm, returning pass/fail with reasoning.

unisound-sm-doc extracts and structures medical entities from clinical notes via internal medical llm, accepting question text or jsonl input and outputting json with extracted fields. deployment is standalone via scripts/run.py.

unisound-med-insurance-fee wraps a medical llm to audit insurance claim documents and calculate reimbursement amounts. input via cli flags or stdin, output structured json with model response and metadata. depends on internal hivoice api key.

unisound-exercise-guidance renders post-operative rehabilitation exercise instructions from structured input, supporting json, csv, and document formats. outputs guided exercise details including name, frequency, duration, and precautions for patient-facing display.

med-followup-record-struct extracts and normalizes chinese outpatient followup records into granular json fields covering history, examination, diagnosis, and treatment recommendations. designed for medical text structuring with privacy controls.

unisound-pe-missing-negative audits outpatient medical records for missing physical examination findings related to diagnosis, delegating judgment to an internal medical llm via hivoice maas api.

unisound-med-record-qc performs quality control on medical records and reports via task-switchable system prompts. supports entity extraction, chart review, clinical pathway validation, and radiology report assessment. single python script deployable standalone.

unisound-health-risk-assessment evaluates multi-dimensional health risks from physical exam reports, stratifies them across cardiovascular, metabolic, tumor and organ domains, and generates time-bound management plans with dietary and lifestyle guidance.

unisound-med-clinical-decision routes clinical queries to an internal medical llm via task-specific system prompts, with support for batch processing and structured json output. integrates disease analysis, primary care triage, treatment planning, and chronic disease management.

detects abnormal chronic disease indicators (glucose, blood pressure, heart rate) against configurable thresholds and generates alert levels with interpretable reasoning. does not diagnose or prescribe treatment.

skill wraps clinical trial descriptive statistics and group comparisons, accepting trial records and endpoint fields to produce structured json output with natural language interpretation via internal medical llm.

tracks post-operative rehabilitation progress by aggregating task completion, pain scores and functional assessments into visualizable metrics and trend analysis. outputs structured summaries and natural language interpretation via internal medical model.

clinic-prescription-assist recommends drug regimens for primary care doctors given patient diagnosis and medication history, checks interactions and contraindications, adjusts dosages for renal/hepatic impairment, and outputs both structured json and natural language summaries.

skill validates outpatient emr records for missing blood pressure and glucose control documentation in patient history by sending deduplicated text to an internal medical llm and returning defect status with reasons.

unisound-lab-report-interpret accepts lab report text (cbc, biochemistry, urinalysis, ekg) and outputs json with flagged abnormalities, clinical significance, and follow-up recommendations. intended for primary care clinicians.

unisound-clinical-trial-design guides structured protocol design for pharmaceutical clinical trials, accepting study parameters like indication, intervention, phase, and endpoints, then producing trial design recommendations via internal medical llm. scope excludes statistics, recruitment, regulatory submission, and drug discovery.

unisound-pre-existing-review routes insurance claim materials through an internal medical llm to classify diagnosis relationships and coverage eligibility, outputting structured audit opinions with no persistence by default.

skill for ranking pharmaceutical target candidates by evidence strength, druggability, and safety risk. accepts disease context and target metadata, outputs prioritized screening results via internal medical llm api.

unisound-followup-reminder generates appointment overdue status and suggests next actions based on last visit date and followup interval. structured around a simple date comparison with optional natural language interpretation via api.

generates initial drafts of academic materials for pharmaceutical medical affairs by assembling topic, audience, key messages and evidence into structured json output. references scientific writing principles but lacks procedural depth and error handling.

unisound-health-education provides a content entry point for patient-facing health knowledge articles, matching topics and keywords against a curated knowledge base, then surfacing relevant materials with disclaimers that education does not replace medical diagnosis.

门诊慢病审核（糖尿病/高血压）。输入 OCR 结果数组 JSON，由内部医疗大模型输出审核结论与原因（原始 JSON + 自然语言结论）。

诊断编码审核。输入结构化病案 record 与待审核诊断列表，输出诊断编码规则命中、病历证据和审核结论。

诊断依据充分性审核。输入结构化病案 record 与待审核诊断列表，输出诊断依据充分性审核结果。

医学名词规范解释。由调用方提供题目文本或含 `question` 字段的结构化输入，经内部医疗大模型按题干约束生成作答；仅含 `scripts/run.py`，可独立拷贝部署。

根据病程记录生成诊疗经过。输入病程记录文本，调用内部医疗大模型，输出结构化诊疗经过文本。

根据住院病历生成标准出院记录。输入入院记录、病程记录等，调用内部医疗大模型，输出结构化出院记录（含主诉、入院情况、入院诊断、诊疗经过、出院诊断、出院情况、出院医嘱七个字段）。

门诊病历内涵质控：主诉不应使用疾病和操作。给定门诊病历文本，调用内部医疗大模型，输出无缺陷或有缺陷及原因。

门诊初诊病历生成。根据医患对话生成结构化的初诊病历信息。

医生端临床科研 — 相似病例语义检索与可解释排序，锚点病例对候选池做类比推理辅助。

医生端临床科研 — 医学文献问答与检索延展，在用户提供的题录/摘要片段上做 PICO 对齐与证据叙事。

门诊病历内涵质控：体格检查结果与现病史不符。给定门诊病历文本，调用内部医疗大模型，输出无缺陷或有缺陷及原因。

检后随访管理。根据体检报告，生成结构化随访计划（时间节点/随访内容/健康指导），支撑体检闭环服务（JSON + 面向受检者的随访通知）。

门诊病历内涵质控：主诉和现病史描述不一致。给定门诊病历文本，调用内部医疗大模型，输出无缺陷或有缺陷及原因。

门诊病历内涵质控：体格检查中遗漏主要阳性体征。给定门诊病历文本，调用内部医疗大模型，输出无缺陷或有缺陷及原因。

病人端慢病管理血糖监测记录。参考 Glucosio 的 blood glucose logging 部分，构建慢病管理基础记录能力。

病人端慢病管理用药提醒。参考 MedTimer 的 medication reminder 部分，构建提升慢病依从性的提醒能力。

病人端通用健康健康档案管理。参考 GitEHR 的 patient-owned health record 部分，构建病史、过敏史等信息管理能力。

病人端通用健康用药记录管理。参考 MedTimer 的 medication list/dose history 部分，构建个人长期用药管理能力。

病人端术后康复功能自测评估。参考 ResearchKit 的 surveys/questionnaires 部分，构建自评量表能力。

门诊慢病审核（糖尿病/高血压）。输入 OCR 结果数组 JSON，输出审核结论与原因（原始 JSON + 自然语言结论）。

将中文门诊复诊病历文本结构化为细粒度字段，输出 JSON（如现病史/既往史/诊断/处理意见等）。

从中文医患对话文本生成门诊初诊病历，输出结构化分段的病历正文（文本）。

病人端术后康复计划查看。参考 CareKit 的 care plan/task model 部分，构建术后康复核心入口。

病人端术后康复今日康复任务。参考 CareKit 的 daily tasks/scheduling 部分，构建日常康复执行任务能力。

药企医学事务管理研发文献分析。参考 Literature Review skill 的 evidence synthesis 部分，构建文献分析能力。

基层常见病诊疗建议。输入患者主诉/症状/体征文本，输出鉴别诊断（概率+依据）、推荐检查、初步处理意见及转诊判断（JSON + 自然语言摘要）。

慢病筛查辅助。输入居民健康数据（体征/检验/生活方式/家族史），评估高血压、糖尿病、心脑血管疾病等慢性病风险，给出分级和管理建议（JSON + 自然语言摘要）。

预防接种提醒。输入居民年龄/接种史/特殊情况，依据国家免疫规划生成个性化接种提醒清单（JSON + 自然语言摘要）。

根据手术相关病历素材生成规范的手术记录。输入术前小结、术中记录、术后记录等，调用内部医疗大模型，输出结构化手术记录（含手术日期、时间、地点、医生、手术名称、指征、诊断、经过、异常情况、术后情况、签名等 14 个字段）。

体检异常指标专项解读。针对体检报告中的每项异常，逐一说明"是什么/为何异常/有何影响/如何干预"，并分析多指标关联，语言通俗易懂（JSON + 专项解读文字）。

复查提醒。从体检报告中提取所有需复查项目，按紧急程度分级（2周/1-3月/3-6月/6-12月），给出复查检查项、就诊科室和准备事项（JSON + 面向受检者的复查提醒）。

体检报告整体解读。输入完整体检报告文本，输出整体健康评级、主要发现（含严重程度）、优先行动清单和个性化生活方式建议（JSON + 通俗易懂的整体解读）。

病人端慢病管理血压监测记录。参考 HealthLog 的 vitals/measurements 部分，构建慢病管理基础记录能力。

健康咨询、体检解读与智能导诊挂号科室推荐。通过 --task 区分咨询与导诊场景；仅含 `scripts/run.py`，可独立拷贝部署。

医学考试、专科问答、文献阅读理解、名词规范解释与术语同义匹配等知识问答。由调用方传入题目，经内部医疗大模型作答；通过 --task 选择具体场景提示词；仅含 `scripts/run.py`，可独立拷贝部署。

临床用药咨询、个性化用药教育与康复指导问答。通过 --task 切换场景；仅含 `scripts/run.py`，可独立拷贝部署。

心理健康疏导与心理卫生知识问答。通过 --task 区分咨询与科普场景；仅含 `scripts/run.py`，可独立拷贝部署。

处方审核（问答/单选/多选题型）、医疗安全规范与医学伦理法规问答；仅含 `scripts/run.py`，可独立拷贝部署。