MyOrbisResults Suite
2025–2026
Lead architect and full-stack developer for MyOrbisResults from 2025 to 2026, responsible for six core brand applications — among them MyOrbisVoice, MyOrbisLocal, and MyOrbisWeb — taken from the initial client concept through to production.
The problem
Six applications under one brand, each with its own surface and audience, all needing to publish content at a volume no editorial team was going to sustain by hand. Treating each as an isolated build would have meant six content pipelines, six schema strategies, and six sets of search assumptions drifting apart within a year.
Approach
- 01
Designed the platform architecture first and derived the applications from it, so the six shared a content model rather than each inventing one.
- 02
Built SEO, AEO, and GEO frameworks into that architecture as a platform concern, not as per-site optimisation work applied afterwards.
- 03
Used multi-agent systems for backend processing, so generation, review, and publishing were separate cooperating steps rather than one prompt doing everything.
- 04
Added custom AI training functionality so model behaviour could be tuned against the platform's own domain rather than left generic.
What was built
- Backend and frontend for six production applications
- Multi-agent systems handling backend processing and automated content generation
- Custom AI training workflows
- Search architecture — classic, answer, and generative — designed into the platform
Stack
- Multi-agent systems
- AI training workflows
- PHP
- JavaScript
- REST APIs
Outcome
Six applications shipped from concept to production, with content generation and search architecture running as platform capabilities rather than per-site manual effort.
In short
Why use multi-agent systems for content generation instead of a single model call?
A single call has to plan, write, check, and format in one pass, and it has no way to fail partially — a bad section means discarding the whole output. Splitting the work across cooperating agents makes each step inspectable and separately correctable: one agent plans structure, another drafts, another reviews against the brief, another handles publishing and markup. Steps can be retried in isolation, different models can be used where their strengths differ, and the review stage can reject work before it reaches production. For content published at volume across six applications, that difference decides whether the pipeline can run without a human reading every output.