网站 ops
代理 01 walks a page audit.
机密提案 · 仅供评估使用
KPMC AI-托管 数字化 平台
Two 专用 AI智能体 for 网站, Medical 日志 & 数字化 基础设施 运营
Powered by AINNA NeuralOps
从静态网站维护迈向持续管理、AI 辅助的数字运营。
KPMC will not simply receive a redesigned website. KPMC will receive an AI-managed digital platform operated by two dedicated AI智能体, supported by AINNA NeuralOps, running on a controlled LAMP and MySQL infrastructure.
2专用 AI智能体
2 / monthIndicative 新 digital modules
连续Automated monitoring capability
Primary demonstration
一个可运行的演示环境，而非生产级医院网站
AINNA 已准备了 KPMC 演示网站，以展示所提议的信息架构、患者就医旅程与期刊体验。该网站仅用于提案与评估。
相关 在线演示 in this proposal
日志 workflow
代理 02 drafts only after sources and review.
Two-agent handoff
已批准 article becomes a website link recommendation.
碳 compare
NeuralOps 与常规网站运营的能耗对比。
This demonstration environment is provided exclusively for proposal and evaluation purposes. It will be removed within seven 天数 following the formal presentation or migrated to an agreed KPMC-controlled domain or environment, where applicable, in consideration of data governance, confidentiality and PDPA requirements.
The demonstration is hosted on an AINNA domain. It is not the permanent production address. 生产 should use a KPMC-authorised domain such as the official hospital domain or an approved subdomain.
关于 AINNA
源自真实运营, then productised
AINNA develops AI-driven operational 系统, NeuralOps 架构, detached 系统, workflow automation and digital platforms. The technology originated from AINNA’s own operating requirements, not as a consultancy slide deck first.
经营 base
AINNA 的内部零售业务涵盖 Shopee、Lazada 与 TikTok 电商平台约 30 家网店，拥有超过 80,000 个 SKU，以及大量产品、电商平台、内容与报表数据。
What that forced
As operations grew, AINNA built AI智能体, detached 系统 and NeuralOps to automate work that would otherwise remain manual, inconsistent and expensive to repeat.
What is offered to KPMC
The same operating discipline — controlled agents, human authority, least-privilege 工具 — 已施加 to a hospital website, journal and LAMP environment.
Key positioning
This Is Not a 网站 维护 Contract
KPMC receives digital operations capability, not merely a website. 传统 arrangements are reactive. AINNA proposes a continuous operating model.
传统 model
Web developer已参与 after a request exists
手动 updatesSomeone notices a problem
更改 request开发人员登录服务器
Periodic maintenance更改 is published, then 空闲
Static website等待下一次投诉
VS
Proposed model
KPMC 管理设定权限、政策与审批流程
AINNA NeuralOpsOrchestrates, routes and governs
AI 智能体 01 + AI 智能体 02Specialised digital operators
网站 + 日志 + 服务器 + Database一个统一平台，而非孤立页面
持续监控 & improvementRecommendations and controlled actions
AINNA 并非提议成为 KPMC 的传统网站管理员。AINNA 提议的是一个 AI 管理的数字运营层。KPMC 仍是最终决策方。
Two-agent operating model
一个治理层，两个专业化运营方。
单一通用型智能体可能造成上下文混杂、审计能力较弱。将网站运营与期刊智能分离，可提升权限管理、故障排查与内容安全。
AINNA NeuralOps
编排 · model routing · validation · tool permission · audit
代理 01
KPMC 网站 运营智能体
企业 website, services, doctors, SEO, navigation, promotions, technical monitoring and website administration.
代理 02
KPMC 日志 智能 代理
日志 workflow, article structure, sources, categorisation, archive, SEO and controlled AI-assisted publishing.
Linux
Apache
PHP
MySQL
文件 storage
网站 CMS
日志 系统
日志
Scheduled tasks
Backups
各智能体如何共享结构化信息
If 代理 02 publishes an approved article on diabetes screening, 代理 01 may recommend linking it from the health screening page, a relevant specialty page, the homepage article module and the internal-link map. That produces a platform, not isolated pages.
为何采用两个智能体而非一个
代理 01 specialises in
网站, SEO, UX, public information consistency and website-related server signals.
代理 02 specialises in
日志, health education, references, editorial states and review cadence.
NeuralOps coordinates both. 隔离 improves governance, permissions, auditability, context quality and troubleshooting.
实时 demo
Two-agent handoff
日志 approved
代理 02 notify
NeuralOps route
代理 01 recommend links
Await KPMC review
就绪. 播放 to see an approved diabetes-screening article proposed for the health-screening page.
AI 智能体 01
KPMC 网站 运营智能体
持续协助 KPMC 面向公众数字形象的管理、组织、优化与技术监控。
网站 content management
监控 and organise hospital profile, service pages, specialist information, doctor profiles, facilities, contact and visiting information, health screening packages, corporate pages, careers, promotions, announcements, events and patient information. The agent flags potentially outdated items and requests review.
内容 consistency
- 部门 spelling variants and title inconsistency
- 复制 service descriptions or pages
- 已过期 promotional dates and old announcements
- 损坏 sections, missing contact details, formatting drift
SEO operations
协助处理标题、元描述、标题层级、内部链接、站点地图一致性、失效链接检测、图片替代文本、内容缺口、关键词覆盖率与重复元数据。这是持续的技术与内容 SEO 工作，但并不保证搜索引擎排名。
用户 experience monitoring
识别失效导航、缺失信息、页面过长、内部链接薄弱、按钮不一致、缺少行动号召、移动端布局问题、重复文案与空页面。
Link monitoring
定期审计内部链接、预约链接、WhatsApp 链接、外部医疗资源、社交账号、PDF、招聘、Qmed 与联系方式链接。
已选择 technical signals
HTTP and application errors, PHP and Apache logs, disk usage, database connection failures, scheduled-task failures, availability, SSL 状态 where accessible, redirects, missing assets, image errors, slow pages and storage growth. This is selected monitoring, not a claim of full 企业 observability.
Proposed public website experience
专业 healthcare navigation such as 首页, 关于 KPMC, Find a Doctor, Medical 服务, Facilities, 健康 Screening, Appointments, 日志, News, 招聘 and 联系. The working demonstration already explores this information architecture.
实时 demo · 代理 01
网站 operations audit
扫描 pages
检查 links
SEO metadata
标记 stale promo
队列 review
就绪. This simulation does not change the 实时 KPMC demo.
AI 智能体 02
KPMC 日志 & Medical 内容 代理
运营结构化的医疗期刊与健康教育出版工作流。这并不是 AI 医生，而是 AI 辅助的出版与知识管理智能体。
KPMC 日志 platform
The journal is a structured knowledge library, not a marketing blog. It should support categories, search, specialties, latest and 精选 articles, archive, publication date, author or reviewer where provided, references, related articles, reading time, medical disclaimer, tags and internal links.
Orthopaedics
O&G
Paediatrics
ENT
一般 medicine
Surgery
健康 screening
Diabetes
Hypertension
Preventive health
Women’s health
Family health
Hospital news
Mental wellness
分类 shown are illustrative and should follow specialties KPMC actually publishes. The demonstration journal already uses a structured category model.
AI-assisted article creation
AI 不会凭空捏造医学声明并发布。所提议的工作流程如下：
已批准 topic
来源 collection
Article draft
Claim validation
Reference check
人类 review
SEO structure
发布 + review cycle
Medical content guardrails
- No diagnosis of individual patients
- No personalised medical advice
- 不编造数据、试验、引文或参考文献
- No unsupported treatment or pharmaceutical claims
- No confidential patient information
- 敏感文章需要人工批准
来源-based content
优先级 sources: KPMC-approved internal information, Ministry of 健康 马来西亚, WHO, peer-reviewed literature, medical society guidance, government health information and KPMC specialist input. 外部 material is not assumed to be legally scrapable or republishable. Use APIs, RSS, licensed sources or manual references.
审核 states and metadata
Each article can carry ID, title, slug, category, tags, draft and publication dates, last reviewed date, author, reviewer, references, agent-generated flag, human-reviewed flag, SEO title, meta description and 状态:
草稿 → AI review → 人类 review → 已批准 → 已发布 → Scheduled review
Article refresh
代理 02 periodically identifies articles that may need revision: older than the agreed review period, broken references, updated guidelines, expired programmes, outdated screening copy or changed specialist details. It recommends review. It does not silently change medically significant information.
实时 demo · 代理 02
Controlled journal workflow
已批准 topic
Collect sources
草稿
验证 claims
人类 review
就绪. The agent will not publish. It stops at human review.
Beyond the front-end
AI智能体 operating the operational foundation
The two agents are not chatbot widgets. They are controlled digital-operations agents that interact with selected server-level and application-level 工具.
AINNA NeuralOps
代理 tool layer
服务器 administration 工具
LAMP stack
网站 + journal applications
MySQL data layer
Linux
Underlying operating environment. Agents may assist with 系统 状态, service checks, storage, file-permission audits, log inspection, scheduled tasks, backup verification, deployment checks and resource usage. 严重 OS changes remain permission-controlled.
Apache
Web server layer. 监控 may cover availability, virtual hosts, HTTP errors, redirects, access and error logs, SSL configuration, URL routing and static asset delivery. Configuration changes are controlled and logged.
PHP application layer
Primary application layer where appropriate. Agents may review error logs, deprecated-function signals, form or API failures, scheduled PHP tasks, file integrity and configuration consistency. 生产 code is never rewritten automatically without governance.
MySQL
Structured data for website content, doctor profiles, services, journal articles, categories, tags, references, SEO metadata, settings, audit logs and agent recommendations. Agents monitor connectivity, table health, size, 失败 or slow queries, duplicates, missing fields, orphans, consistency and backup 状态.
Database access model
必需
AI 智能体 → permission layer → validated tool/API → MySQL. 最小权限. No unrestricted destructive access by default.
Avoided
AI 智能体 → unrestricted root database credentials. Controlled 工具 are safer than giving an LLM raw database access.
AINNA NeuralOps
编排 and governance, not uncontrolled autonomy
NeuralOps is the layer that controls how AI智能体 interact with the website, server, data and external models.
KPMC user
NeuralOps
Classify
Retrieve
Select model
验证
许可 tool
执行 · verify · log
模型 routing
不同的任务并不需要相同的模型。NeuralOps 可根据复杂度、隐私、成本、速度、上下文长度、推理与编码需求进行路由。该架构可支持云端 LLM、开源模型、本地模型与专业模型。本提案并不绑定单一供应商。
独立系统
Not every task should 通过 through a large model. 确定性 系统 handle stable work: database validation, broken-link scanning, sitemap generation, backup checks, article scheduling, metadata validation and uptime checks. AI is used where language, interpretation or decision support is required. That reduces token use, cost, latency, hallucination exposure and external-model dependency.
知识 hierarchy
- KPMC-approved structured database
- KPMC-approved documents
- 已批准 medical reference sources
- 一般 LLM knowledge, last
可信度更高的受控来源优先于通用模型知识。
Reducing hallucination through architecture
幻觉 cannot responsibly be described as eliminated. 风险 is reduced through controlled sources, retrieval, structured databases, validation rules, human approval, tool restrictions, output checking, logging, agent separation and detached 系统. This proposal does not claim “zero hallucination”.
ESG and compute efficiency
Use rules, scripts, database queries, 轻量模型s, specialised agents, cached structured data and detached 系统 before escalating to a larger model. That can reduce unnecessary compute and token consumption. No carbon-reduction figure is claimed here.
Future local AI
该架构仍与未来本地或开源模型保持兼容，凡商业与技术条件适宜之处均可：数据控制、更低的 API 依赖、更可预测的成本、专业模型与本地部署选项。这并不意味着每个模型都会在 KPMC 内部运行。
权威性 and control
AI operates the 系统 — KPMC retains authority
智能体协助持续运营。KPMC 保留对医疗内容、企业信息、医生信息、定价、促销、临床信息、公开声明及患者相关政策的决策权。AINNA 在约定权限范围内管理数字基础设施与自动化层。
权限 model
| Level | The agent may | The agent may not |
|---|---|---|
| L1 观察 | Read website 状态, logs, content, database metadata and SEO data | Modify anything |
| L2 推荐 | 准备 recommendations, draft content and suggested corrections | 发布 without approval |
| L3 Controlled action | 更新 approved text, publish approved articles, update metadata, repair low-risk issues — all logged | 更改 medical or corporate facts without policy |
| L4 Restricted admin | 提议架构、Apache、PHP、安全或服务器配置变更 | 未经授权的技术批准不得执行 |
内容 approval matrix
低风险
SEO 元数据、失效链接、格式、图片优化、技术修复。可在政策允许下自动化处理。
中等 risk
服务 descriptions, hospital announcements, promotions. 商业 approval depending on policy.
高 risk
Medical advice, treatment information, clinical claims, medication content. 已授权 review required.
PDPA and healthcare data
The public website should minimise handling of sensitive patient medical information. If future 系统 process personal data, apply PDPA controls: data minimisation, consent, retention, controlled access, 审计追踪s and secure transmission. This proposal does not create a clinical patient-data 系统 unless separately approved.
患者-data separation
Public website & journal
医院信息、教育内容、预约界面。
Hospital clinical 系统
HIS / EMR remain isolated. Future integration only through controlled APIs and approved interfaces. Clinical databases are not exposed to the public website or to AI智能体.
安全 model
最小权限, role-based access, credential separation, secrets kept out of prompts, encrypted communications, access logging, database permission separation, production/staging separation, backup protection, rate limiting, input validation, secure uploads, PHP hardening, sanitisation, SQL-injection prevention, XSS and CSRF protection. No certification is claimed unless independently held.
数据 ownership
KPMC should retain ownership of KPMC content, doctor information, medical articles, hospital data and website data generated for KPMC. AINNA owns its proprietary NeuralOps 架构, agent framework, automation technology and generic 系统 components, subject to contract.
数字化 operations
连续 数字化 运营
监控
理解
推荐
批准
执行
验证
学习
监控
更改 management
AI identifies an issue → recommendation → authorised review → approved change → agent executes → 系统 validates → audit log. That loop is the operating difference versus a ticket-and-wait webmaster.
日志记录 and audit
Important actions record timestamp, agent ID, user, task, action, tool, data affected, previous and 新 value, approval 状态, result and error 状态.
备份 strategy
生产 → daily application backup → database backup → encrypted off-server storage → retention policy. 推荐 daily logical database backups, periodic full backup, recovery testing and backup-log monitoring. 保留 periods are set with KPMC, not assumed here.
Environments and deployment
Development → staging → production. 高-impact changes test in staging first. 部署: validate → backup → staging test → approval → production → post-deploy health check.
可用性 and incidents
Agents may monitor HTTP response, homepage and journal availability, MySQL connectivity, PHP and Apache errors, disk space and key pages. On error: analyse logs, classify severity, attempt only approved low-risk recovery or escalate to the AINNA technical team, then record the incident. AI cannot automatically resolve every server incident.
AINNA technical team
智能体并不会消除人的技术责任。AINNA 仍负责在约定服务范围内维护并改进架构。智能体是团队的延伸，而非替代。
AI-enhanced, not AI-dependent
If AI is unavailable, the website continues, the journal remains readable, and booking links continue to function. 严重 website operations must not depend on an LLM being online.
推荐 architecture
KPMC users → security layer if applicable → Apache → PHP → MySQL (website data + journal data), alongside 代理 01, 代理 02, NeuralOps and the controlled tool layer. 可选 integrations: Qmed, WhatsApp, analytics, CRM, email and other approved APIs. Unconfirmed components are labelled as recommended, not as already deployed.
Qmed
The current KPMC environment references Qmed. Proposed phases: (1) direct booking link, (2) embedded experience where technically and contractually permitted, (3) API 集成 if Qmed provides suitable APIs and KPMC approves. API availability is not claimed without confirmation.
AI-enhanced search — future capability
Natural-language questions such as “Which doctor should I contact for knee pain?” should guide users to relevant specialties and information. The 系统 must not diagnose the patient.
Future management dashboard
代理 01
网站 agent · tasks · alerts
代理 02
Drafts · reviews · publications
服务器
CPU · RAM · disk
Database
状态 · backup · size
网站
正常运行时间 · broken links · SEO findings
安全
警告 · 失败 logins · updates
仪表盘 cards are a proposed future management interface, not a claim that a 实时 hospital operations console is already in production for KPMC.
分析
潜在 monitoring includes page views, popular services, doctor-profile visits, appointment CTA clicks, journal traffic, search queries, navigation paths, device mix and referrals — subject to privacy and consent requirements.
Reporting
Periodic operational reports can cover website updates, agent activity, journal activity, technical alerts, SEO findings, broken links, content recommendations, module development, security events, backup 状态 and server health.
商业 continuity
数据库、文件与配置备份、恢复流程、监控、智能体日志以及文件化的手动回退方案。
实施
参考路线图——需经范围审批
第一阶段 — 基础
Finalise website, production environment, LAMP, MySQL, backup, 代理 01, 代理 02 and permission model.
第二阶段 — 内容 migration
企业 content, services, doctors, facilities, news and journal — from approved KPMC sources only.
Phase 3 — 代理 activation
监控, content auditing, SEO, journal workflow, logging and approval controls.
Phase 4 — 集成
潜在 Qmed, WhatsApp, analytics, CRM and email — each subject to technical and contractual confirmation.
Phase 5 — 连续 development
根据 KPMC 的优先级，每月约两个已批准的数字模块。不是十个。除非正式立项，否则不保证交付。
可选 future agents
代理 03 appointments/enquiry, 代理 04 marketing, 代理 05 analytics, 代理 06 internal knowledge. 新谜题 agents can be added without redesigning the platform.
每月两个新的数字模块
参考候选模块（非承诺清单）：医生搜索、预约入口、健康筛查搜索、期刊、医疗常见问题、招聘、活动、促销、专科名医目录、患者与访客指南、保险合作机构目录、套餐对比、企业媒体中心、健康计算器、电子简报、患者咨询、WhatsApp 入口、CRM 集成、数据分析面板。
服务器 ownership options
A — AINNA-managed
更快的支持与集成的智能体管理，部署更简单。
B — KPMC-controlled
KPMC 拥有托管。智能体在受控访问下运行。更强的内部自主权。
C — 混合
KPMC controls production. AINNA maintains development/staging and agent 系统. Often the better hospital-governance fit. No contractual choice is made in this document.
服务 governance
KPMC 管理 → KPMC digital / marketing / IT representative → AINNA technical director / project team → AI智能体. Separate escalation paths for content, medical, technical, security and integration issues.
What AINNA is responsible for
AI 智能体 platform, NeuralOps, website technology, journal platform, agreed server configuration, LAMP environment, MySQL application layer, agent tooling, automation, continuous development, technical monitoring and 系统 optimisation.
What KPMC controls
医疗政策与批准、医生信息、医院政策、公开声明、临床内容审批、企业决策与患者信息政策。
Commercial positioning
此处不标注价格。商业范围取决于托管安排、集成需求、模块数量、SLA、安全要求、培训、支持与部署模式。
ESG · compute efficiency
碳足迹：NeuralOps 与常规网站运营对比
本节估算了以下各项的算力能耗与 CO₂e： managing a hospital website and journal — audits, drafts, SEO, link checks, logs — not the carbon of every public page view. 图表 use the same layer model as the AINNA 碳模拟器.
What is compared
传统: most operational tasks are sent to a large language model. NeuralOps: detached 系统 and rules handle scans, backups and validation; a model is used only when language or judgement is required.
Grounded factors
电网 factor default 0.74 kg CO₂e/kWh, PUE 1.4, and kWh per 1,000 requests by layer — all defaults from the AINNA 碳模拟器. Token reduction of up to 87% is an 内部基准 on a tested language workload, not a hospital-site measurement.
What this is not
Not a certified carbon audit. Not a claim of KPMC’s actual emissions. Not a guarantee of 87% reduction on every task. Adjust the sliders; the model recalculates 实时.
—
— kWh
—
— kWh
—
— kg CO₂e / month
— kg / year
≤87%
内部 benchmark on tested token workload — 已施加 only as context, not multiplied into the kg figure.
| Layer | 对网站运营的意义 | kWh / 1,000 tasks | 传统 share | NeuralOps share |
|---|---|---|---|---|
| GPU 密集型 AI | 完整 LLM for every rewrite, scan summary or log read | 0.15 | 70% | 5% |
| 轻量 AI / CPU | Short classification or title suggestion | 0.05 | 20% | 15% |
| 规则-based | 验证, metadata, schema, spelling lists | 0.01 | 8% | 20% |
| 分离式 系统 | 链接爬取、站点地图、备份检查、可用性探测 | 0.005 | 2% | 60% |
估算 / simulation only. Formula: tasks × layer share × (kWh per 1,000 tasks) × PUE × grid factor. 来源: AINNA 碳模拟器 defaults. 更改 any input to see sensitivity. Do not treat the result as audited hospital ESG data.
Final message
KPMC 不 Need Another Static 网站.
KPMC can operate a continuously evolving digital platform managed by specialised AI智能体, 受治理的 by humans and supported by AINNA NeuralOps.
AINNA proposes a transition from conventional website maintenance to an AI-assisted digital operations model. Two specialised AI智能体 will support KPMC’s website, medical journal, LAMP server environment and MySQL data layer while operating within defined permissions, governance controls and human approval processes.
其结果不仅仅是重新设计的网站，而是一个旨在与 KPMC 共同持续演进并发展的数字运营平台。
The website is the interface. The journal is the knowledge platform. The LAMP server is the operational foundation. MySQL is the structured data layer. The two AI智能体 are the digital operators. NeuralOps is the orchestration and governance layer. KPMC remains the authority.
