TeachAdapt
The lesson adapts. The learner never has to.
Mainstream e-learning assumes a sighted, hearing, English-reading student on broadband. TeachAdapt is designed to begin at the opposite end — the learner as they are, on the phone they hold, in the language they think in — and to reshape the lesson without diluting it. Today it drafts lessons as text in the web chat, and nothing is delivered by WhatsApp, SMS or voice call.
- Channels (designed)
- Web todayPlanned: WhatsApp · SMS · Voice
- Languages (designed)
- English · French · Arabic · Swahili · Spanish · Hindi
- Designed for
- Global · Africa · Asia · Pacific
- Tools
- 11 declared5 not connected · 6 illustrative
Begin with Asha
An illustrative portrait, not a real learner: Asha is fourteen. She has low vision and a mild intellectual disability, attends an under-resourced school in Kenya’s Turkana County, and shares one feature phone with three siblings. The curriculum is in English and Swahili; she is fluent in Turkana. Mainstream e-learning is a twenty-megabyte JavaScript bundle her phone cannot render and her family cannot afford the data for. TeachAdapt’s system prompt opens with her; everything below answers to her.
“Disability, low literacy, low connectivity, and unfamiliarity with technology are constraints on the system around the learner, not deficiencies in the learner.”
Channel before content
TeachAdapt is designed to choose how to reach a learner before choosing what to say — channel constraints govern length and form, never the reverse — starting at the lowest tier the phone supports, ranked by its Low-Bandwidth Delivery Patterns skill. Of the five rungs below, only the web exists today. The other four are designed behaviour: those channels are not deployed.
USSD — designed, not deployed
A 182-character menu, six numbered options, ninety seconds — on every phone ever sold.
SMS — designed, not deployed
160 characters in GSM-7, seventy in Arabic or Hindi script — counted before sending.
Voice (IVR) — designed, not deployed
Short sentences, key facts repeated. Press 1 to repeat, 0 for a human — in the planned deployment; no human is on call today.
WhatsApp — designed, not deployed
Plain text, no markdown — and never sent without a consent token.
Web — the only channel today
Aimed at WCAG 2.2 AA: real headings, alt text, 4.5:1 contrast, keyboard-operable. Not audited yet.
The same lesson, four ways
Prose can only assert adaptation; a specimen shows what is meant. Below is a single sixty-word lesson — why clean hands matter — written out the four ways TeachAdapt is designed to render it for declared access needs. Move between the tabs: the meaning holds; only the form moves.
Germs are very small.You cannot see germs.Germs can make you sick.Soap washes germs away.Wash. Count to twenty.Wash before you eat.Wash after the toilet.Clean hands keep you safe.
- Hands
- Soap
- Water
- Twenty
- Food
Germs are tiny living things. They are far too small to see, but they can make you sick. Germs sit on your hands and travel to your food, your mouth, and other people. Soap breaks germs apart so water can wash them away. Count slowly to twenty while you wash — before you eat, after the toilet, and after touching animals.
Lesson: why clean hands matter. One section, three list items. Germs are living things too small to see. They travel from hands to food, to mouths, and on to other people. Soap breaks germs apart so water can rinse them away. Wash for twenty seconds, at three moments. One: before eating. Two: after using the toilet. Three: after touching animals. End of list. End of lesson.
- [Sign: GERM TINY-TINY]Germs are tiny living things.
- [Sign: SEE CANNOT]You cannot see them.
- [Sign: HANDS GERM SIT]Germs sit on your hands.
- [Sign: SICK CAN]They can make you sick.
- [Sign: SOAP WATER WASH]Soap and water wash germs away.
- [Sign: COUNT TWENTY]Count to twenty while you wash.
- [Sign: EAT BEFORE, TOILET AFTER]Wash before eating and after the toilet.
Rule appliedReading age about nine: short sentences, concrete verbs, one idea each. The full curriculum point survives intact.
This switch illustrates the design of the adaptive_tutor_step tool: one lesson plan, re-rendered to the learner’s declared access needs — equivalent access, never a reduced version, per the ITU Accessibility Guidelines skill. It is not the tool’s output: the lesson and its four renderings were written by hand for this page, and no signing avatar exists yet.
A short glossary of access
Four terms from TeachAdapt’s skills explain the agent better than any feature list.
- POUR
- Perceivable, operable, understandable, robust — the WCAG 2.2 AA Compliance skill’s four tests for every screen-rendered lesson.
- The 5Rs
- Retain, reuse, revise, remix, redistribute — the Mozilla Open Education Principles test: block any one and it is, at best, free to look at.
- Warm digital handover
- The Good Things Foundation Digital Inclusion pattern for moving a learner between programmes: a named person on both sides, never a cold referral.
- Digital Champion
- A peer who finished the modules and helps the next cohort — matched on language, lived experience and region, only with mutual consent.
Pedagogy from named frameworks
The pedagogy is transcription, not improvisation. The UNESCO Inclusive Education Framework skill supplies the test every lesson must pass — presence and participation: delivered but inaccessible counts as marginalised. The ITU Accessibility Guidelines skill supplies the equivalent-access rule the specimen above enacts. And the Critical AI Literacy Curriculum skill is written to teach learners to question, contest and audit AI itself — recognition, bias awareness, provenance, agency.
What it will never do
Ten system-prompt rules are non-negotiable. Five define the agent:
- It never patronises. It matches the learner’s intellect and adapts the channel.
- It never infers a disability. Access needs are self-declared; it asks once at most, and treats nothing as declared until the learner confirms it.
- It is designed never to message without a consent token, and never to share a champion’s contact unless both sides opt in. Today it sends no messages and matches no champions.
- It never accepts a cognitive-function test as the only way in — WCAG 3.3.8, flagged wherever found.
- It never bluffs. Below 0.80 confidence it says it is unsure; in the planned deployment, a named human is asked to look. A safeguarding signal halts tutoring at once.
“A confident wrong answer is the worst failure mode; an honest ‘I don’t know, here is a human who does’ is the right one.”
Whose frameworks it follows
No affiliation or endorsement. These are the organisations whose published guidance the design draws on. No partner system is connected.
The design draws on published work: UNESCO’s curriculum frameworks; the ITU’s channel and accessibility standards; Mozilla’s open resources and critical-AI curriculum; Good Things Foundation Australia’s champion model. The agent works from that guidance as written into its skills, not from any of their systems.