LMS 2.0: a learning platform with a module-scoped AI tutor · Case study

Problem

This institution runs on a specialised body of knowledge: infrastructure and distributed-renewable-energy project finance, ESG safeguards, credit structuring. New staff and partners have to learn it, and it mostly lived in people’s heads and scattered documents. LMS 2.0 makes it teachable, trackable, and answerable in context.

Onboarding into complex project-finance workflows was slow and informal.
Reference material (SOPs, glossary, FAQs) lived somewhere else, so learners kept leaving the lesson to find it and losing their place.
Managers had no view of who had learned what.
Course owners got no structured feedback they could act on.

Evidence / research

The platform has three audiences with different questions. A learner asks “what does this term mean, and what’s the procedure?”, usually in the middle of a lesson. A manager asks “who on my team has covered what?”. A course owner asks “which parts of this module aren’t landing?”. The old setup answered none of them in place: definitions and SOPs lived in other documents, progress lived in nobody’s view, and feedback, when it came at all, was too vague to act on.

My role

I designed the learner and admin experiences, the AI assistant interactions and contributions to the design system shared with the CRM.

Constraints

The design had to be delivered in two months, for content covering specialised infrastructure-finance knowledge. The assistant needed a clearly communicated module boundary, and the interface needed to serve learners, managers and course owners within the shared design system.

Options considered

The default for an AI assistant is a general chatbot: it answers anything, which looks impressive in a demo and is risky in credit structuring, where a confident wrong answer does real damage. The default for reference material is a link out to a shared drive, which is cheap and keeps documents in one place, but pulls the learner out of the lesson every time. And a single interface for everyone is less to build, but it buries the learner’s next step under course-management controls they will never use.

Decision

Scope the AI to the module, and say so. A general chatbot invites off-topic questions and confident wrong answers about finance. Scoping it to the current module, and labelling that limit clearly, trades breadth for trust, which is the correct trade when the subject is credit structuring.

Bring reference material into the module. SOP and glossary tabs mean the learner never abandons the lesson to look something up.

Split learner and course-owner experiences behind a role switch, rather than one crowded view that serves neither.

Make feedback structured and answerable. Multi-dimension ratings plus reply threads give course owners something specific to act on.

Design

What I did

Designed a phased course model: curricula split into stages (Origination, Due Diligence, …), each module carrying its own Content, Materials, SOPs, Glossary and FAQ tabs, an assessment, and a certificate on completion. Seven surfaces inside every module, so nothing sends the learner elsewhere.
Designed a module-scoped AI assistant: “ask anything about this module”, with suggested prompts (summarise, key risks, define a term) and a visible “scoped to this module” label, grounded in the module content, not the open internet.
Added a manager dashboard, a discussion forum, structured multi-dimension course reviews with reply threads, and course-owner reports.

LMS course detail with phased curriculum — Course detail: phased curriculum, outcomes, resources

LMS module player with resource tabs — Module player: Content, Materials, SOPs, Glossary, FAQ in one place

LMS module-scoped AI assistant — The assistant, labelled 'scoped to this module', with suggested prompts

LMS module assessment with scored questions and a pass or fail result — Each phase ends in a scored assessment before the learner moves on

LMS certificate of completion — A certificate on completion, tied to the programme

LMS team dashboard — Manager view: who has learned what

Collaboration / implementation

I worked within the DREEF design team, covering the learner and admin experiences and the assistant’s interaction design. Because the LMS shares a design system with the CRM, components I contributed serve both products, so the two tools feel like one suite to the staff who use them.

Measurement

Onboarding time, completion rates, assistant usage, and whether learners read the “scoped” limit as relief or restriction, are still early. I hold them as targets rather than confirmed results, with firm figures to follow once there is a stable stretch to report.

Result

Every module now carries seven reference surfaces, content, materials, SOPs, glossary, FAQ, feedback and the assistant, so a learner never has to leave the lesson to find an answer. The AI assistant stays scoped to the module in front of the learner rather than the open web, and the platform shares one design system with the CRM.

What I learned

Designing the assistant made its boundaries part of the core experience. The module label, suggested prompts and nearby reference material all help explain what a learner can ask. That boundary needs to remain understandable throughout the conversation, including when a question falls outside the module.

My next priority would be testing whether learners understand that scope and can return to the lesson after getting help. I would also look at the questions the assistant cannot answer: they may point to missing course material as well as opportunities to improve the assistant itself.