What you’ll build and why
The problem
A group with several companies needs every page in two languages to say the same true thing. Keeping that true by hand means re-checking translations, facts and pages one at a time. Automating it without controls is worse: a model can draft a page that repeats a topic you already cover, states a claim no source supports, or goes live in one language without its twin. Publishing is only the start, because over time a page can slip, overlap with a newer page or lose its independent value. Content at that scale needs evidence before prose, checks before publication, a person who signs off, and measurement after the page is live.
What you’ll build
This Build Guide walks you through a reference SEO automation system for evidence-grounded bilingual content. Scheduled jobs discover search demand and compare it with your existing pages, so each opportunity becomes a deterministic decision to create, refresh, merge or reject. Sources are gathered and stored before any writing starts, and one evidence-backed brief produces native English and Arabic drafts that carry the same supported facts. Every revision then runs four automated gates (claims, quality, SEO and publication policy) and waits for an approver to sign off before it is published. After publication, the system verifies the live page and keeps measuring it with Search Console and indexation data. It queues a refresh, merge, redirect or noindex action only when its thresholds and rollback checks pass, and that change also waits for sign-off. A small state machine and a scheduler that advances one bounded step at a time let interrupted work resume safely.
Who it’s for
The guide is for developers, technical leads and content or SEO teams who run sites in English and Arabic, including across several companies or entities, and who want automation that stays accountable, with every factual statement in both languages tied to a source. The named stack is GitHub for version control, Postgres for workflow and evidence state, OpenRouter as a server-side model gateway, and a Web Hoster / Cloud Infrastructure for the application and its schedules. All of them are educational choices you can replace, and every record and identifier in the guide is fictional. It does not describe Ting’s private systems, it leaves out Ting’s prompts, scoring rules and internal thresholds, and it does not guarantee rankings, traffic or revenue.
What this educational system demonstrates
This reference describes a reusable pattern for evidence-grounded bilingual content operations. Scheduled jobs discover demand, compare existing pages, gather verifiable sources, draft both locales, run four independent automated gates, hold every passing revision until an approver signs off, then publish it and keep measuring each page. The records and identifiers are fictional examples, and the design is not a description of Ting’s private systems.
Included
- Demand discovery with existing-content deduplication
- Evidence records linked to every factual clause
- Paired English and Arabic revisions
- Approver sign-off before publication, then automatic verification, monitoring and lifecycle signals
Not included
- Ting’s production architecture or provider choices
- Private prompts, scoring rules or internal thresholds
- Real customer, analytics or credential data
- A guarantee of rankings, traffic or revenue
Resumable architecture and state machine
A small state machine defines legal transitions, while a scheduler advances one bounded step per invocation. Durable storage holds leases, evidence, decisions and retries so interrupted work resumes safely. This is one educational pattern; teams should adapt the states and limits to their own risk, volume and platform.
- 1ClaimsEvery fact linked to evidence
- 2QualityClear, specific, no duplicate intent
- 3SEOTitle, links, canonical, structured data
- 4PublicationBoth languages paired; policy decision recorded
- Passon to the next gate
- Repairback to drafting, bounded retries
- Rejectstops; nothing publishes
Refresh cycle: back to evidence
Source
stateDiagram-v2
[*] --> DISCOVERED
DISCOVERED --> ELIGIBILITY_CHECK
ELIGIBILITY_CHECK --> CLUSTERING
CLUSTERING --> RESEARCHING
RESEARCHING --> EVIDENCE_READY
EVIDENCE_READY --> BRIEF_GENERATING
BRIEF_GENERATING --> DRAFTING
DRAFTING --> CLAIM_VALIDATION
CLAIM_VALIDATION --> QUALITY_VALIDATION
QUALITY_VALIDATION --> SEO_VALIDATION
SEO_VALIDATION --> AWAITING_APPROVAL: all four gates pass
AWAITING_APPROVAL --> READY_TO_PUBLISH: approver signs off
READY_TO_PUBLISH --> PUBLISHING
PUBLISHING --> PUBLISHED
PUBLISHED --> MONITORING
MONITORING --> REFRESH_REQUIRED: decay detected
REFRESH_REQUIRED --> RESEARCHING: refresh cycleSource
stateDiagram-v2
CLAIM_VALIDATION --> REPAIRING: unsupported claim
SEO_VALIDATION --> REPAIRING: gate failed
AWAITING_APPROVAL --> REPAIRING: changes requested
REPAIRING --> DRAFTING: bounded retries
DRAFTING --> FAILED_RETRYABLE: provider error
FAILED_RETRYABLE --> DRAFTING: backoff resumeNamed reference stack
The example uses GitHub for version control, Postgres for workflow and evidence state, OpenRouter as a server-side model gateway, and a Web Hoster / Cloud Infrastructure for the application and schedules. These are educational reference choices, not claims about Ting’s production stack.
GitHub
Reference version control and CI for code, policy fixtures and deployment history.
Web Hoster / Cloud Infrastructure
Application runtime, scheduled triggers, secret injection, health checks and rollback.
Postgres
Illustrative durable store for topics, evidence, immutable revisions, decisions and lifecycle signals.
OpenRouter
An educational server-side model gateway choice with schema-constrained output, timeouts and fallback policy.
Scheduled Jobs + Application Functions
The reference runs discovery and worker schedules at configurable cadences; each tick performs one bounded step and reserves time to persist safely.
Setup, code, tests and deployment are in the PDF
This page shows the scope, the architecture and the reference stack with its official documentation. The implementation sections listed below are in the full Build Guide PDF: enter your name and mobile number once and it downloads. This device then stays unlocked for every free Ting resource.
- 04 · Repository, schedules and setupIn the PDF
- 05 · Data and state contractsIn the PDF
- 06 · Eligibility and clusteringIn the PDF
- 07 · Research and claim groundingIn the PDF
- 08 · Bilingual draftingIn the PDF
- 09 · Quality and publication gatesIn the PDF
- 10 · Monitoring and lifecycleIn the PDF
- 11 · Scheduled API and failure behaviourIn the PDF
- 12 · Testing and evaluationIn the PDF
- 13 · Deployment and operationsIn the PDF
- 14 · Security and exclusionsIn the PDF