Australian SCHADS Award compliance — in minutes, not days.澳大利亚 SCHADS Award 合规检查 —— 分钟级而非天级。
Cognitive Reading × Rules Knowledge Base认知阅读 × 规则知识库
Production deployment · SCHADS Pay Guide, NES, LSL Act ingested 生产部署 · 已摄取 SCHADS Pay Guide、NES、LSL Act
SCHADS Award is dense. Classification, minimum wages, penalty rates, allowances, overtime, super, leave loading — payroll teams check each clause by hand against every role and every roster. SCHADS Award 条款繁杂。分类、最低工资、罚率、津贴、加班、养老金、年假加成 —— 工资团队对每个岗位、每张排班手工核对每条条款。
Underpayment is silent. Wrong classification, missing Sunday loading, forgotten 11.5% super — none of these throw an error. They just compound, sometimes for years. 欠薪问题不会自动报警。分类错误、周日加成漏算、11.5% 养老金遗漏 —— 这些不会触发任何报错,只会年复一年累积。
Every change re-opens the check. New hires, role transitions, roster swaps, Award updates — each event can invalidate a prior compliance call. 每次变动都要重新核对。新员工入职、岗位变更、排班调整、Award 更新 —— 任意事件都可能使之前的合规结论失效。
Voluntary disclosure rarely catches everything. Fair Work cases routinely uncover gaps the original audit missed — by then the back-pay window is years wide. 主动补报难以穷尽。Fair Work 案例屡屡发现原审计遗漏的差异 —— 此时追溯窗口已是数年。
Result: HR sits on quiet liability. The system below replaces the manual check with a structured, citable analysis. 结果:HR 团队背着隐性责任。下文系统用结构化、可溯源的分析替代手工核对。
Upload SCHADS Pay Guide, NES, LSL Act PDFs. The system splits, indexes and stores them as a navigable knowledge graph — searchable by clause, role and concept. Already in production. 上传 SCHADS Pay Guide、NES、LSL Act PDF。系统自动分块、索引、入库,形成可按条款/岗位/概念检索的知识图谱。已生产部署。
Organisation-specific rules (e.g. internal allowances, EBA overrides) added through the Knowledge Base UI — no restart, no code change. 机构特定规则(如内部津贴、EBA 覆盖)通过 Knowledge Base UI 添加 —— 无需重启、无需改代码。
Tier 1 — Rules checklist: classifications, minimum wages, Saturday 150%, Sunday 200%, Public Holiday 250%, overtime, super 11.5%, leave entitlements. 第 1 层 —— 规则清单:分类、最低工资、周六 150%、周日 200%、公共假日 250%、加班、养老金 11.5%、休假权益。
Tier 2 — Methodology: how to identify role, extract scenario facts, retrieve relevant rules, score compliance, format the output. Same methodology applies to every case. 第 2 层 —— 方法论:如何识别岗位、提取场景事实、检索相关规则、判定合规性、格式化输出。每个案例使用同一套方法。
The LLM reasons over retrieved clauses + analysis methodology — explanations cite specific Award sections, not generic prose. LLM 基于检索到的条款 + 分析方法论推理 —— 解释直接引用具体 Award 条款,而非泛泛而谈。
doc_rules/ for new PDFs and adds them to the database; click Ingest on each to process.
One-time setup. Upload the SCHADS Pay Guide, National Employment Standards (NES), Long Service Leave Act — once each. The cognitive reader splits, indexes, and stores them in MySQL.一次性配置。上传 SCHADS Pay Guide、国家就业标准(NES)、Long Service Leave Act —— 各一次。认知阅读器自动分块、索引、入 MySQL。
Each clause becomes searchable. When a case analysis fires, the system retrieves only the clauses relevant to that role + scenario — citations are anchored to the original PDF text.每条条款均可检索。运行案例分析时,系统仅检索与该岗位 + 场景相关的条款 —— 引用锚定到原文 PDF。
doc_rules/knowledge_history/; analysis automatically uses the latest.
Award 更新了?重新上传新版 PDF 即可。旧版归档到 doc_rules/knowledge_history/,分析自动使用最新版。
Edit organisation-specific rules in plain text. EBA top-ups, internal allowance rates, role-specific exceptions — written in natural language under [USER_ADDITIONS_SECTION].用纯文本编辑机构特定规则。EBA 加价、内部津贴标准、岗位特殊例外 —— 用自然语言写在 [USER_ADDITIONS_SECTION] 下面。
Save → immediately effective. No restart, no code, no deployment cycle.保存 → 立即生效。无需重启、无需编码、无需部署。
Each case = one role + one roster. The Cases page lists every active case. Click any to open its documents and start analysis.每个案例 = 一个岗位 + 一份排班。Cases 页面列出所有进行中的案例,点击任意一项即可打开文档、开始分析。
The demo case CASE-41 SAIL-demo is preloaded with a Family Day Care Educator role — PD + Scenario already attached.演示案例 CASE-41 SAIL-demo 已预装家庭日托教育者岗位 —— 含 PD + 实际工时文档。
Two documents per role. The PD describes what the role should do; the Scenario records what actually happened — hours, shifts, allowances paid.每个岗位两份文档。PD 描述岗位应当的职责;Scenario 记录实际发生的工时、班次、津贴支付。
Format-agnostic: PDF, DOCX, TXT — the system parses each and pairs them up automatically.格式不限:PDF、DOCX、TXT 均可,系统自动配对解析。
Every answer cites the exact Award clause. Citations link back to the ingested SCHADS / NES / LSL knowledge graph — no black-box response.每个回答都引用具体的 Award 条款。引用链回已摄取的 SCHADS / NES / LSL 知识图谱 —— 不存在黑盒回答。
Chat history saved per case. Ask broad questions or drill into specific shifts, allowances, or interpretive edge cases.聊天记录按案例保存。可以问宽泛问题,也可以追问特定班次、津贴或解释性边界场景。
## Overall Assessment
The scenario contains multiple calculation errors. Jason's base hourly rate ($32.35/hr) does not match any classification level in the ingested SCHADS table. Superannuation uses 11% instead of the standard 11.5%. Higher-duties calculations for Level 4 pay point 1 are inconsistent with the Award. Annual leave accrual is correct. Parental leave responses align with NES.场景中存在多处计算错误。Jason 的基础时薪 $32.35/hr 在已摄取的 SCHADS 表中找不到匹配等级。养老金按 11% 计算而非现行 11.5%。Level 4 第 1 工资点的高级别工作计算与 Award 不一致。年假累积正确。育儿假回复符合 NES。
## Findings
The report is a structured markdown response, rendered in the chat. H2 sections (Compliance Status, Overall Assessment, Findings) plus per-finding H3 with severity tag [High] [Medium] [Low], Details and Calculation sub-blocks.报告是结构化 Markdown 回复,直接在聊天中渲染。H2 章节(Compliance Status、Overall Assessment、Findings),每条发现 H3 + 严重度标签 [High] [Medium] [Low],含 Details 与 Calculation 子段。
5 streamed steps shown in chat: chunk → extract → search KB → compare → generate. Each step appears live as the pipeline runs.5 个流式步骤实时显示在聊天中:分块 → 提取 → 检索知识库 → 比较 → 生成。每步随流水线推进实时展现。
doc_chat_analysis_report (MySQL). Continue chatting in the same case to drill into any finding — the report stays as context.
留作记录。存入 doc_chat_analysis_report(MySQL)。在同一案例继续聊天可针对任意发现追问 —— 报告作为上下文保留。
A working SCHADS demo case (FDCE PD + Scenario) is pre-loaded. Log in, open CASE-41 SAIL-demo, and run the analysis end to end. FDCE 演示案例(PD + 实际情况)已预装。登录后打开 CASE-41 SAIL-demo,完整跑一遍分析流程。
Prof. Yuan Miao · Intelligent Technologies Australia 苗原教授 · 澳大利亚智能技术公司
yuan.miao@aits.softwareAgentCenter.com
Production deployments in healthcare (Polyps), compliance (SCHADS) and education (KidSmart). Cognitive reading pipeline reused across all three. 已在医疗(Polyps)、合规(SCHADS)与教育(KidSmart)三个领域生产部署。认知阅读流水线在三者间复用。
Built on 26 years of research in cognitive computing & fuzzy logic. 基于 26 年认知计算与模糊逻辑研究积累构建。