网站 ops
代理 01 walks a page audit.
Confidential proposal · Evaluation use only
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 static website maintenance to continuously managed AI-assisted digital operations.
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
A working demonstration environment, not a production hospital site
AINNA has prepared a KPMC demonstration website to illustrate the proposed information architecture, patient journey and journal experience. It exists solely for proposal and evaluation.
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 vs conventional website-ops energy.
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 systems, NeuralOps 架构, detached systems, 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’s internal retail operations have involved approximately 30 online stores across Shopee, Lazada and TikTok commerce, with more than 80,000 SKUs and large volumes of product, marketplace, content and reporting data.
What that forced
As operations grew, AINNA built AI智能体, detached systems 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 requestDeveloper logs into the server
Periodic maintenance更改 is published, then 空闲
Static websiteWaits for the next complaint
VS
Proposed model
KPMC 管理Sets authority, policy and approvals
AINNA NeuralOpsOrchestrates, routes and governs
AI 智能体 01 + AI 智能体 02Specialised digital operators
网站 + 日志 + 服务器 + DatabaseOne platform, not isolated pages
持续监控 & improvementRecommendations and controlled actions
AINNA is not proposing to become KPMC’s conventional webmaster. AINNA is proposing an AI-managed digital operations layer. KPMC remains the authority.
Two-agent operating model
One governance layer. Two specialised operators.
A single general-purpose agent creates mixed context and weaker audit. Separating website operations from journal intelligence improves permissions, troubleshooting and content safety.
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
How the agents share structured information
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.
Why two agents instead of one
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
To continuously assist with the management, organisation, optimisation and technical monitoring of KPMC’s public-facing digital presence.
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
Assists with titles, meta descriptions, headings, internal linking, sitemap consistency, broken-link detection, image alt text, content gaps, keyword coverage and duplicate metadata. This is continuous technical and content SEO. It is not a guarantee of search rankings.
用户 experience monitoring
Identifies broken navigation, missing information, excessively long pages, weak internal links, inconsistent buttons, missing calls to action, mobile layout problems, repeated copy and empty pages.
Link monitoring
Periodic audit of internal links, appointment links, WhatsApp links, external medical resources, social profiles, PDFs, recruitment, Qmed and contact links.
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
To operate a structured healthcare journal and health-education publishing workflow. This is not an AI doctor. It is an AI-assisted publishing and knowledge-management agent.
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 does not invent medical claims and publish them. The proposed workflow is:
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
- No fabricated statistics, trials, quotes or references
- No unsupported treatment or pharmaceutical claims
- No confidential patient information
- Sensitive articles require human approval
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
Execute · verify · log
Model routing
Different tasks do not require the same model. NeuralOps may route by complexity, privacy, cost, speed, context size, reasoning and coding need. The architecture can support cloud LLMs, open-source models, local models and specialised models. This proposal is not locked to a single provider.
独立系统
Not every task should 通过 through a large model. 确定性 systems 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
Higher-confidence controlled sources take priority over generic model knowledge.
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 systems. 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 systems before escalating to a larger model. That can reduce unnecessary compute and token consumption. No carbon-reduction figure is claimed here.
Future local AI
The architecture remains compatible with future local or open-source models where commercially and technically appropriate: data control, lower API dependency, more predictable cost, specialised models and on-premise options. This does not imply that every model will run inside KPMC.
权威性 and control
AI operates the 系统 — KPMC retains authority
Agents assist with continuous operation. KPMC retains authority over medical content, corporate information, doctors’ information, pricing, promotions, clinical information, public statements and patient-related policies. AINNA manages the digital infrastructure and automation layer within agreed permissions.
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 | Propose schema, Apache, PHP, security or server configuration changes | Execute without authorised technical approval |
内容 approval matrix
低风险
SEO metadata, broken links, formatting, image optimisation, technical fixes. May be automated under policy.
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 systems 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
Hospital information, education content, appointments interface.
Hospital clinical systems
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
Execute
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
Agents do not remove human technical responsibility. AINNA remains responsible for maintaining and improving the architecture 在 agreed service scope. Agents extend the team; they do not replace it.
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
Database, file and configuration backups, recovery procedures, monitoring, agent logs and a documented manual fallback.
实施
Indicative roadmap — subject to scope approval
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
Approximately two approved digital modules per month, according to KPMC priorities. Not ten. Not guaranteed deliverables unless formally scoped.
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.
Two 新 digital modules every month
Indicative candidates, not a committed catalogue: doctor finder, appointment gateway, health screening finder, journal, medical FAQ, careers, events, promotions, specialist directory, patient and visitor guides, insurance panel directory, package comparison, corporate media centre, health calculator, newsletter, patient enquiry, WhatsApp gateway, CRM integration, analytics dashboard.
服务器 ownership options
A — AINNA-managed
Faster support and integrated agent management. Simpler deployment.
B — KPMC-controlled
KPMC owns hosting. Agents operate with controlled access. Stronger internal ownership.
C — 混合
KPMC controls production. AINNA maintains development/staging and agent systems. 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
Medical policy and approval, doctor information, hospital policies, public statements, clinical content approval, corporate decisions and patient-information policies.
Commercial positioning
No price is stated here. Commercial scope depends on hosting arrangement, integration requirements, number of modules, SLA, security requirements, training, support and deployment model.
ESG · compute efficiency
碳 print: NeuralOps vs conventional website operations
This section estimates the compute energy and CO₂e of 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 systems 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 internal benchmark 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 | What it represents for website ops | 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% |
| 独立式 系统 | Link crawl, sitemap, backup check, uptime probe | 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.
The result is not merely a redesigned website. It is a digital operating platform designed to evolve continuously with KPMC.
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.
