Opaque spend
Programmes could not prove learning impact; engagement died early.
SK Assist
· AISatyakabir Assistant
Satyakabir Technologies
Technology engineering · Digital transformation
Work · Case study
Adaptive learning for 800,000 students across six countries
800K
learners
+94%
course completion*
6
countries live
8 mo
to multi-region GA
Case study
We built an adaptive LMS with AI tutoring, multilingual delivery, and analytics that finally connected programme spend to learning outcomes — 800K learners, +94% course completion on redesigned paths.
Challenge
Content was localised by hand, engagement collapsed after week two, and leadership could not prove which programmes worked. Instructors lacked tools; finance lacked evidence.
Solution
An adaptive learning core with mastery models, AI tutoring under pedagogical guardrails, multilingual delivery pipelines, and analytics that instructors and finance could both trust.
How we framed the work
Programmes could not prove learning impact; engagement died early.
Adaptive paths and analytics designed with instructors, not around them.
Tutoring that helps without inventing unsafe or off-curriculum answers.
Six countries live with completion lift on redesigned paths.
Engagement arc
Mastery model & goals
LMS core & paths
Guarded AI assist
Six-country delivery
Analytics & GA
Cadence
Mastery model, metric definitions, MVP scope.
Paths, instructor tools, first locale.
Guarded AI, warehouse views.
Remaining locales, enablement, handoff.
01 · Context
Headcount of learners rose while completion fell. Localisation was artisanal. Finance asked which courses deserved budget; nobody had a trustworthy answer. Instructors needed intervention tools, not vanity dashboards.
02 · Architecture
We separated the mastery graph from CMS content so pedagogy could evolve without CMS rewrites. Events streamed into a warehouse with instructor and finance views. The AI tutor sat behind curriculum constraints and offline evals before any cohort saw new behaviour.
03 · Delivery
Each locale wave included instructor councils. Accessibility and keyboard paths were acceptance criteria. Localisation pipelines replaced hand-copy chaos. Analytics definitions were agreed with finance before charts shipped.
04 · Safety & integrity
Tutor prompts refused to complete graded work. Assessment modes isolated when required. Student data residency followed country agreements. Abuse and harassment reporting shipped in the learner surface.
05 · Outcomes
800K learners across six countries. Redesigned mastery paths showed about +94% completion vs prior comparable courses*. Instructors used intervention queues weekly. Finance finally had programme-level outcome views tied to spend.
06 · Transfer
We transferred authoring tools, eval harnesses, and analytics ownership. Satyakabir remains available for new locales and model upgrades under the same pedagogical gates.
01
Learner and instructor web experiences
02
Adaptive path engine with mastery gates
03
AI tutor with pedagogical guardrails
04
Localisation pipeline for six locales
05
Institutional analytics and export packs
Mastery graph separate from content CMS
Tutor suggestions evaluated offline before promote
Event stream into warehouse for instructor + finance views
WCAG-oriented UI with keyboard and screen-reader passes
From the engagement
Adaptive learning for 800,000 students across six countries
Stack & standards
“We finally see which programmes work — and learners stay long enough for mastery to matter.”
Continue exploring