Step 01
AI Automation Audit
Find the workflow worth automating first.
Maps bottlenecks, handoffs and automation suitability before anyone commits to a build.
Workflow mapOpportunity scorePilot recommendation
Maliwan Savage | AI Architect
Agentic AI systems for content and media publishing
I design practical AI operating layers that turn messy manual processes into governed workflows with context engineering, agent orchestration, retrieval grounding, and human checkpoints.
Content and media publishing workflowsSEO and localisation systemsInternal AI toolsWorkflow auditsPilot builds
Services
Choose the engagement path that matches your needs.
Step 02
Pilot Build
Turn one high-value workflow into a working system.
Designs the states, validation rules, retries and human checkpoints needed for daily use.
Working prototypeValidation gatesOperator handoff
Step 03
Advisory Support
Strengthen internal AI systems already in motion.
Gives senior input on architecture, delivery risks, tooling choices and production readiness.
Architecture reviewRisk inputDelivery guidance
Case Studies
Proof systems, not long project notes.
Link governanceSEOML retrievalHuman reviewPublishing controls
Governed Internal Linking
Stage 1
Link candidates
Stage 2
Eligibility checks
Stage 3
SEO review
Stage 4
Controlled publish
Client need
Help SEO teams find useful internal links without sending readers to the wrong market, intent or destination.
System designed
Developed an ML-assisted matching algorithm using multilingual Vertex AI embeddings and dual-signal passage retrieval to rank candidate links by keyword and destination-page similarity. LLM intent checks and approved-anchor, market and destination rules gate human review, exact diffs and reversible publishing.
Operational value
Resolved SEO-reported audit errors and reduced unresolved destination mappings in a defined audit cohort.
Review queueExact link diffUndo trail
Link governanceSEOML retrievalHuman reviewPublishing controls
Governed Internal Linking
Stage 1
Link candidates
Stage 2
Eligibility checks
Stage 3
SEO review
Stage 4
Controlled publish
Client need
Help SEO teams find useful internal links without sending readers to the wrong market, intent or destination.
System designed
Developed an ML-assisted matching algorithm using multilingual Vertex AI embeddings and dual-signal passage retrieval to rank candidate links by keyword and destination-page similarity. LLM intent checks and approved-anchor, market and destination rules gate human review, exact diffs and reversible publishing.
Operational value
Resolved SEO-reported audit errors and reduced unresolved destination mappings in a defined audit cohort.
Review queueExact link diffUndo trail
Read all 8 case studies in full
Case study 01
Governed Internal Linking
Turns link opportunities into reviewable, controlled publishing decisions.
SEOML retrievalHuman reviewPublishing controls
Workflow flow
- 1Link candidates
- 2Eligibility checks
- 3SEO review
- 4Controlled publish
- Client need
- Help SEO teams find useful internal links without sending readers to the wrong market, intent or destination.
- System designed
- Developed an ML-assisted matching algorithm using multilingual Vertex AI embeddings and dual-signal passage retrieval to rank candidate links by keyword and destination-page similarity. LLM intent checks and approved-anchor, market and destination rules gate human review, exact diffs and reversible publishing.
- Operational value
- Resolved SEO-reported audit errors and reduced unresolved destination mappings in a defined audit cohort.
Outputs
- Review queue
- Exact link diff
- Undo trail
Case study 02
SERP Driven AI Content Engine
Turns SERP signals into clear writing direction.
SEORAGSERP intelligence
Workflow flow
- 1SERP signals
- 2Retrieval context
- 3Brief generation
- 4Optimisation loop
- Client need
- Find ranking opportunities, analyse competitors and convert reliable research into publishable editorial direction without fragmented tools.
- System designed
- RAG retrieval, SERP analysis, competitor comparison, keyword clustering, source-grounded briefs, writing assistance and optimisation loops.
- Operational value
- Improves SERP intelligence, research quality, content velocity and strategic content prioritisation.
Outputs
- SEO brief
- Topic cluster
- Writing direction
Case study 03
Brand-governed Creative Production Engine
Brand-safe visuals for multi-site editorial operations.
Brand governanceCreative opsPublishing
Workflow flow
- 1Brand profile
- 2Image workflow
- 3SEO metadata
- 4Publish routing
- Client need
- Create brand-safe featured images, infographics and SEO-ready media without ad-hoc prompts or inconsistent handoffs.
- System designed
- Brand profiles, reference-image guidance, generation/editing, SEO metadata, WordPress publishing and a Brand Governance API.
- Operational value
- Improves creative throughput, brand governance, asset provenance and reusable downstream automation.
Outputs
- Creative asset
- Featured image
- Governed API
Case study 04
AI News Localisation & Publishing
Turns source news into localised, SEO-ready CMS drafts.
LocalisationWordPressEditorial workflow
Workflow flow
- 1Source news
- 2Localisation
- 3Review gate
- 4WordPress target
- Client need
- Translate, enhance, review and route news across markets without manual research, prompt writing and publishing steps.
- System designed
- Source extraction, AI-assisted localisation, SEO targeting, enrichment, featured images, side-by-side review and validation.
- Operational value
- Increases localisation velocity, editorial quality control, publish safety and revenue enablement.
Outputs
- Localised draft
- SEO metadata
- Publish-ready content
Case study 05
AI Music Agentic System
Creates song concepts, lyrics and scored output candidates.
AgenticCreative IPPrompt governance
Workflow flow
- 1Intent
- 2Agent roles
- 3Self-audit
- 4Candidate scoring
- Client need
- Produce high-quality catalogue ideas without manual prompt trial-and-error, weak hooks or repeated themes.
- System designed
- Composition planning, lyric writing, vocal fit checks, sonic profiles, self-audit, scoring and export payloads.
- Operational value
- Improves creative throughput, prompt governance, catalogue scalability and candidate quality.
Outputs
- Lyrics
- Style prompt
- Music candidate
Case study 06
Doc-to-CMS Draft Orchestration
Converts structured source documents into predictable CMS drafts.
DocumentsCMSValidation
Workflow flow
- 1Source document
- 2Field mapping
- 3Validation
- 4Draft creation
- Client need
- Move faster from source document to publishable CMS draft without repeated copy-paste and formatting cleanup.
- System designed
- Document parsing, field mapping, validation, bounded retries and operator checkpoints before publish.
- Operational value
- Creates faster drafts, less formatting drift and clearer visibility into workflow state.
Outputs
- CMS draft
- Mapped fields
- Workflow state
Case study 07
PR Translation & Multi-site Publishing Pipeline
Multi-market release workflow that localises source content and routes validated drafts across sites.
TranslationMulti-siteRelease workflow
Workflow flow
- 1Source release
- 2Local rules
- 3Draft routing
- 4Validation
- Client need
- Coordinate multilingual rollout when translation, localisation and publishing steps are fragmented.
- System designed
- Source extraction, localisation rules, prompt-controlled translation, routing and required-field validation.
- Operational value
- Improves multilingual turnaround, consistency across locales and launch reliability.
Outputs
- Localised release
- Validated draft
- Launch-ready copy
Case study 08
Slack-triggered Workflow Orchestration
Internal workflow system that turns Slack requests into routed tasks with explicit state handling.
SlackInternal toolsRouting
Workflow flow
- 1Slack request
- 2Intent mapping
- 3Owner routing
- 4Completion signal
- Client need
- Reduce unclear ownership, missed steps and inconsistent follow-through from chat-based operational requests.
- System designed
- Intent mapping, owners, downstream stages, validation gates, retries and escalation for unresolved states.
- Operational value
- Reduces coordination overhead and makes ownership, next actions and completion signals visible.
Outputs
- Routed task
- Escalation state
- Clear owner
Credibility
Built from real operational work, not demo theatre.
15person
SEO function led
Operational leadership across content, publishing and measurable delivery.
200%
organic revenue growth
Commercial grounding for choosing automation that moves real outcomes.
Multi-site
publishing experience
Hands-on work across translation, routing, release discipline and AI-assisted delivery.
Operational leadership
Workflow design shaped by ownership, measurable outcomes and day-to-day operating pressure.
Commercial and technical grounding
MBA background, MIT professional training and practical implementation across APIs and automation.
Production delivery mindset
Focus on trusted systems: explicit states, validation, checkpoints and clear handoffs.
AI Automation Profile
When to hire Maliwan for AI automation and internal productivity systems.
Maliwan Savage designs governed AI workflow automation for teams that need reliable execution, not isolated prompts. She turns messy operational processes into practical agentic AI systems with clear inputs, retrieval grounding, orchestration, review steps and measurable output quality.
Roles she matches
Maliwan is best matched to work described as AI automation architect, AI automation specialist, AI automation consultant, AI workflow architect, AI transformation builder or AI systems consultant.
- Useful when a team needs a builder who can translate operational pressure into working internal tools.
- Strong fit for teams that need practical judgement across content, publishing, SEO, localisation and AI operations.
Problems she solves
She focuses on repeated workflows where manual handoffs, unclear ownership, inconsistent prompts or disconnected tools slow delivery.
- Workflow audits that identify which process should be automated first and which should stay human-led.
- Pilot builds that prove value before a team commits to a wider AI workflow automation programme.
Systems she builds
Her systems use AI operating layers that connect models, retrieval grounding, orchestration, validation and human checkpoints around real work.
- Agentic AI systems with explicit states, retries, review gates and production-ready handoffs.
- Internal AI productivity systems for research, briefing, translation, creative production, routing and CMS drafting.
Workflows and sectors
The strongest proof sits in SEO, localisation, content and media publishing, where speed only matters if quality, brand fit and publish safety hold.
- SEO and SERP research engines, localised editorial workflows and multi-site WordPress publishing pipelines.
- Internal tools for Slack-triggered routing, document-to-CMS orchestration and repeatable operational requests.
Contact
Bring one workflow. Get a practical AI route.
Send the workflow, bottleneck, users and whether you need an audit, pilot or senior input.
Best first message
What should change?
A short note is enough: the workflow, the bottleneck, the users and the outcome you want to make repeatable.
Audit a manual workflow
Scope a focused pilot
Review an internal AI build
Technical Stack
Practical stack by operating layer.
Layer 1
Models
The model layer covers selecting and steering LLMs for retrieval, reasoning, drafting, scoring and review tasks. It includes context engineering so model output is grounded in useful source material instead of loose prompt guesses.
- Model selection
- Context engineering
- RAG reasoning
- Output scoring
OpenAIGPT Sol 6.1Opus 5.5Vertex AICodex
Layer 2
Orchestration
The orchestration layer turns prompts into controlled workflows with prompt contracts, explicit workflow states, validation gates, retry controls and human checkpoints before anything important is published or handed off.
- Prompt contracts
- Workflow states
- Validation gates
- Retry controls
n8nPrompt contractsWorkflow statesRetry controls
Layer 3
Interfaces
The interface layer gives operators clear tools for submitting requests, reviewing generated work, comparing source and output, approving drafts and routing tasks through APIs or Slack integrations.
- Operator review
- API workflows
- Slack requests
- Approval screens
TypeScriptNext.jsReactPythonSlack integrations
Layer 4
Publishing
The publishing and infrastructure layer connects validated outputs to real destinations, including WordPress REST API publishing, API handoffs, deployment on Vercel, source control in GitHub and services hosted on Railway.
- WordPress REST API
- API handoffs
- Vercel deployment
- GitHub and Railway
WordPress REST APIAPIsVercelGitHubRailway