The classroom was a compromise.
One room, one pace, one instructor talking at the speed of the median — because for two hundred years that was the only way to teach thirty people at once. It is not the only way any more.
Lumen authors a programme once, instantiates it per cohort, runs the weekly delivery engine, gives every learner an AI mate grounded in your own material — and emits the evidence pack as a query instead of a scramble.
Interface design · synthetic demonstration data
Training is delivered on spreadsheets — and nobody finds out it failed until it has.
A funded learnership, a graduate intake, a corporate go-live curriculum and a compliance refresher look like four different problems. They fail in exactly the same four ways.
No live view
Progress lives in attendance registers and a facilitator's head. By the time a status report is compiled, the window to intervene has already closed.
Support goes to whoever asks loudest
The learner quietly falling behind — attending, silent, slipping — is the one who fails. They are also the one nobody is looking at.
Evidence is assembled by hand
Compliance and funder reporting is rebuilt at the end from spreadsheets that were never designed to be audited.
The record stops at the door
Certification is captured. Whether the person was actually placed, and is still in the role a year later, is not.
Every one of those is a data problem wearing a training costume. Lumen treats the programme itself as data — authored once, versioned, instantiated per cohort, and executable.
Point it at a qualification. Get a programme back.
This is what makes Lumen a platform rather than one programme with a login screen. Feed it an occupational qualification, a unit-standard set, your existing courseware or a stack of process documents.
In
Qualification specification, assessment criteria, existing decks and exercises, SOPs, product and configuration documentation.
Parse
Chunked and indexed — becoming the same corpus Lumi later answers your learners from.
Propose
Phases, modules, sequence, typed activities, an assessment blueprint mapped criterion by criterion, and the evidence each one requires.
Gate
A person reviews, edits and publishes. Nothing goes live on a model's say-so, and unmapped criteria block publication.
Out
A versioned template. Cohorts bind to it permanently, so you can always prove what a learner was assessed against.
Interface design · synthetic demonstration data
Feed it what you already have.
Two pipes in. Nothing gets retyped, and nothing sits in a folder called "course material" that only one person can find.
A spreadsheet becomes a live cohort in minutes
- 01Roster in, any shape
From a spreadsheet, an existing LMS export or an API. The model reads your column headers and proposes the mapping; a person confirms it.
- 02Validated on the way in
Duplicate detection, identity check, consent capture and invite dispatch — before anyone is enrolled, not after an audit.
- 03Profiled at the start
A baseline assessment plus background and preference, so the first week is pitched at the person rather than at the average.
- 04Grouped deliberately
Study groups formed from the baseline, so a group is a mix rather than an accident of the alphabet.
Material becomes structure, not a document dump
- 01It tells you what it read
Outcomes, knowledge and practical modules, credits — and how much of your existing courseware is actually reusable.
- 02It tells you what is missing
Assessment criteria with no corresponding content are flagged, and publication is blocked until they are authored. The system says what it cannot do.
- 03Timing falls out of it
Instantiate a template against a start date and a roster and the whole calendar materialises — phases, weeks, activity due days, publish cadence.
- 04The corpus is the same one
What you ingested to build the programme is what Lumi answers from. There is no second, unmanaged pile of content.
Everyone reaches the same standard. Nobody has the same week.
My Week is the product.
Monday to Friday, one column a day, each card a target with its type, its weight and its tick. Today is highlighted. There is nothing to interpret and nowhere to hide.
Interface design · synthetic demonstration data
Dated commitments
Targets published every Sunday for the week ahead. An explicit, dated commitment per learner is the unit that has never existed before.
Group work is first-class
Group tasks and peer reviews appear on every member's board with the lead named — so a group carried by one person shows as skew rather than as a pass.
Credentials in the open
What is achieved, what is in flight and what is booked, against the milestone it belongs to.
Points, streaks, league
Group leagues rather than a public individual leaderboard. Points mirror credentials and never gate them.
A tutor each, at eleven at night, that has read everything you gave it.
Not a chatbot bolted to the side of a course. Lumi is pinned to the learner's current position in the programme — it knows which module, which activity, and what they got wrong last week.
One-to-one support has always been the thing that works and the thing nobody can afford. This is the first version of it that scales past one facilitator per twelve people.
- 01Explain
The same idea five different ways until one of them lands.
- 02Drill
Practice against the outcome, not trivia.
- 03Walk through
Step by step through a process, at the learner's pace.
- 04Mark
Open answers against the rubric, with the gap named.
- 05What next
The next target — not the whole plan, which is how people freeze.
Answer from your corpus and your approved sources. Correct a learner up to a consultant-grade answer rather than just marking them right. Offer the drill that fixes the specific thing they got wrong.
Answer from the open internet. Supply exam dumps. Or invent something when the corpus does not cover the question — it says so, escalates to a human, and that deferral is logged as a content gap for the next version.
Support becomes a management signal.
Every conversation is tagged as it is written — node, topic, intent, frustration, and whether Lumi answered from the corpus or deferred. That tagging is what produces the confusion heatmap, the who-needs-help list, and next intake's content backlog.
A learner who is confused asks. A learner who is drowning goes quiet. No questions and no completed targets is the most dangerous signal in the system — and it is precisely the one an attendance register cannot see. Lumen is instrumented to catch it.
Who needs a human this week — generated, not guessed.
Interface design · synthetic demonstration data
The cohort grid
Every group's completion by week. A track sliding from green to amber across three weeks is visible as a shape long before it is visible as a result.
The confusion heatmap
Modules against topics, sized by question volume and coloured by re-asked within 24 hours — the strongest signal an explanation did not land. That is Monday's re-teach list.
The check-in list
Ranked from status and chat signals with the reason attached: quiet and behind, re-asking one concept, a skewed group, a frustration flag raised at 21:40.
The view nobody has ever had: the middle of the programme, live.
For every rand committed — how many learners are on track to complete, certify and be placed. Today, not in a report about last year.
Interface design · synthetic demonstration data
Does it adapt to the learner? Yes — and deliberately not everywhere.
This is the most common question, and the honest answer is a selling point to a moderator rather than a limitation. Two layers, kept strictly apart.
Pace. Sequence within a week. Remediation depth. Extra drills. Which support arrives when. How Lumi pitches an explanation. What the facilitator re-teaches on Monday. All driven by assessment trajectory, status and chat analytics — because it changes the support, never the standard.
The curriculum, outcomes and assessment criteria are fixed. A system that quietly rewrites the syllabus mid-cohort is not innovative, it is unauditable, and it would fail the first external moderation it met. Improvements flow into the next version through a human gate.
Reporting becomes a query, not a scramble.
Every stage writes an audit event as it happens, so the compliance pack is extracted from the running programme rather than reconstructed from it afterwards.
Same engine. Different template.
Each of these is a programme with a fixed spine, a delivery layer that should adapt, a support burden that does not scale, and someone who has to prove it happened.
Accredited & funded training
Providers running occupational qualifications and learnerships. The compliance burden is the strongest argument, and the market is being forced to re-map its offering inside a regulatory window it did not choose.
ERP implementation enablement
Every ERP programme carries a training workstream invented from scratch, delivered once, and evaporating at go-live. Your process design becomes the corpus; end users get a mate that knows your build, not the vendor's manual.
Corporate onboarding
New-hire ramp, role certification, product enablement. Measured in time-to-competence rather than courses completed, with the same early warning on who will not make it.
Graduate & early-career
Structured intakes with a technical spine, group work, a capstone and a placement outcome — the whole journey from enrolment to alumni inside one record.
Franchise & field force
Distributed teams, high turnover, mandatory standards, and a head office that has to prove consistency across sites it cannot visit. The exception queue is the entire value.
Public-sector capacity building
Large intakes across delivery partners of visibly different quality, and a funder that currently discovers failure a year late. Portfolio visibility is the buying reason.
What changes for you — and what never changes for anyone.
The engine is fixed. Everything you would reasonably expect to be yours, is yours: the programme, the corpus, the roles and approval gates, the branding and terminology, the integrations, and where the data sits.
What never changes: the human approval gate on curriculum, the immutable version binding a cohort to what it was assessed against, and the refusal to answer outside the corpus. Those three are the product. Remove any of them and what is left is a content-management system with a chat window.
| Capability | Status |
|---|---|
| Programme engine templates, versions, instantiation, weekly targets | Live |
| Learner portal & Lumi My Week, modules, groups, credentials, the AI mate | Live |
| Facilitator desk queue, cohort grid, heatmap, marking | Live |
| Attendance & audit trail captured in-platform, timestamped, attributable | Live |
| Studio — curriculum compiler qualification in, proposed programme out | In build |
| Funder console portfolio, grants, tranches, exception queue | In build |
| Identity assurance & verified presence | In build |
| Statutory & funder returns configured per jurisdiction and per funder | In build |
| Outcome tracking past the exit placement, then 3 / 6 / 12 months | Roadmap |
We publish this table because the rest of this site would be worth less if we did not. Ask us on the day and you will get the same answer.
Running a programme somebody has to verify?
A learnership, an internship, a graduate intake, a go-live curriculum, or internal capability building — if someone downstream needs proof it happened, that is the conversation.