Building a multi-tenant, AI-powered learning platform for scalable course delivery
Executive summary
Running a learning platform for more than one brand or client from a single codebase is a common ambition and a common failure point: without the right architecture, "multi-tenant" quietly becomes "one install per client," and every update, fix or compliance change has to be repeated by hand. That risk, maintenance cost and inconsistency compounding with every new tenant onboarded, is the business problem this project was built to remove.
Babini Mazzari restructured the platform's core around a multi-tenant architecture, layered in AI-assisted content generation for slides, quizzes and audiovisual material, and hardened the environment for production use. The engagement ran across multiple milestones and a broad set of delivery workstreams, closing with a formal handover to the client's own team.
- Client
- a technology provider building a white-label, multi-tenant e-learning platform for other organisations to run their own branded course catalogues
- Platform and technologies
- WordPress, WP Freighter (hybrid mode), Tutor LMS, OnlyOffice (PPTX editing), SCORM/H5P, Video.js, Wasabi cloud storage, Akool AI avatars, AI content generation pipeline
At a glance
| Before | After |
|---|---|
| A learning platform with no scalable path across brands | A multi-tenant architecture serving multiple branded tenants from one shared codebase |
| Manual content authoring for every course | An AI-assisted pipeline generating slides, quizzes, audiovisual material and SCORM/H5P exports |
| A single environment with no separation between development and production | Staging and production environments split, with dedicated demo environments for sales |
| Video delivery dependent on third-party AI video generation | A proprietary video upload path with Wasabi cloud storage and download-link regeneration |
Business context
For any organisation running a learning platform across multiple brands, departments or client accounts, the strategic question is rarely "can we build an LMS"; it is whether the platform can scale without scaling cost and risk in lockstep. A single-tenant or loosely-shared setup means every new site multiplies support effort, security surface and the chance of brands drifting out of sync with each other. For a CIO or COO sponsoring this kind of work, the real deliverable is not a feature; it is a platform that keeps those variables flat as the business adds tenants.
The client wanted to move from a single-brand learning product to a platform other organisations could adopt as their own, without rebuilding the underlying technology for each one. That meant rethinking the core architecture, rebuilding the content-creation workflow around AI rather than manual authoring, and reaching a stability level suitable for paying tenants rather than a single internal use case.
Challenges
- Restructuring without disrupting existing content. The LMS core had to support multiple tenants from a single codebase without breaking course content already in place.
- Automating pedagogical content creation. Slides, quizzes, audiovisual material and SCORM/H5P exports all had to be generated through an AI-assisted pipeline while preserving instructional quality and a controlled revision flow.
- Integrating multiple AI and document services. OnlyOffice for PPTX editing, Akool for AI avatar generation, and a proprietary video upload path all had to work as one coherent authoring workflow.
- Reaching production-grade stability. The platform needed separate staging and production environments, tenant-level quotas, security hardening and demo environments ready for live sales conversations.
Key decisions and trade-offs
Proprietary video upload instead of third-party AI video generation. The team replaced a dependency on AI-generated video with a proprietary upload path backed by Wasabi cloud storage. The trade-off was giving up automated video creation in favour of full control over quality, storage costs and download-link lifecycle.
Evaluate a process orchestrator, then choose not to build one. An orchestrator for content-generation workflows was assessed and deliberately not implemented, because the existing pipeline was already structured enough to manage without an additional abstraction layer. The trade-off was accepting some manual sequencing in exchange for lower complexity and maintenance.
Hybrid WP Freighter architecture with separate wp-content. The team optimised WP Freighter to run in hybrid mode with separated wp-content directories, giving each tenant its own content boundary without a full multisite rebuild. The trade-off was a more complex deployment configuration in exchange for tenant isolation at a lower infrastructure cost.
Approach
Platform restructuring
The LMS core was restructured to support multiple tenants from a single codebase. WP Freighter was optimised for hybrid-mode stability with separated wp-content directories, giving each tenant a content boundary without a full multisite rebuild.
AI content optimisation and slides
An AI-assisted content pipeline was built around slide templates with a structured revision flow. OnlyOffice was integrated as an online PPTX editing tool, and the prompting pipeline was optimised to produce consistent, reusable slide content.
Exports, quizzes and learning formats
SCORM and H5P export capabilities were added, with the SCORM player improved through Video.js integration. Quiz authoring was introduced for formal course delivery, and subtitles for accessibility were added to the video pipeline.
Production readiness and handover
Staging and production environments were split. Tenant-level quotas were defined to control resource usage per tenant. A proprietary video upload path replaced the AI video generation dependency, backed by Wasabi cloud storage with download-link regeneration. A Wordfence security hotfix on Tutor LMS was applied. Demo environments with dedicated configurations were stood up for commercial use, including a branded demo with a dedicated slide template. A usage guide for AI content generation was produced, and the platform was handed over to the client's team.
Outcomes
A production-ready, multi-tenant LMS
The platform moved from a single-tenant setup to a multi-tenant architecture serving multiple branded tenants from one shared codebase, with tenant isolation, quotas and dedicated demo environments.
AI-assisted content pipeline
A manual content-authoring process was replaced with an AI-assisted pipeline covering slides, audiovisual material, quizzes and SCORM/H5P export, with OnlyOffice for PPTX editing and Akool for AI avatar generation.
Cloud storage and environment separation
Dedicated Wasabi cloud storage and a staging/production split gave the platform a stable foundation for scaling content and tenants, with a proprietary video upload path replacing third-party AI video dependency.
Demo-ready for tenant sales
Demo environments with dedicated configurations and branded slide templates equipped the business with live sales tools to onboard new tenants.
Who this is relevant for
This approach suits EdTech operators, training providers, franchises and groups with multiple customer-facing brands that need to run a single learning platform across tenants without multiplying operational overhead for every new site.
FAQ
What does "multi-tenant" mean for an LMS platform?
It means multiple brands, clients or departments are served from one shared platform and codebase, with each tenant's content, branding and data kept logically separate, instead of running a separate installation per tenant.
Can AI really generate course content that is ready for learners?
AI can produce slides, quizzes and audiovisual material as a first draft, but a controlled revision flow with human review is essential. The pipeline accelerates authoring; it does not replace instructional design.
Why replace AI video generation with a proprietary upload path?
AI-generated video depended on a third-party service with limited control over quality, cost and availability. A proprietary upload path with Wasabi storage gave the team full control over the video lifecycle, from encoding to download-link regeneration.
What happens at project handover?
The platform, documentation, usage guides and demo environments are delivered to the client's team. Ongoing support is a separate arrangement, not an automatic continuation.
Related services
- AI Consulting: designed and integrated the AI-assisted content generation pipeline.
- Application modernization: restructured the LMS core for multi-tenant architecture.
- Web Development: built and stabilised the platform across staging and production.
- Cloud Computing: configured Wasabi cloud storage and environment separation.
Next steps
If your organisation is weighing a multi-tenant rebuild or extension of an existing LMS, a short architecture review is the lowest-risk way to see what a transformation like this would involve for your platform.
