Building Kilele Archade: an AI Learning Platform for Kenya's CBC Curriculum
Why build an EdTech product in Kenya
The CBC competence-based curriculum asks students to explore, and teachers to guide across large classes with limited resources. Generic chatbots do not know the CBC syllabus, a Kenyan school's notes or a student's actual competence level. I wanted an AI tutor grounded in the material students are actually examined on — built here in Nairobi, shipping to real classrooms.
Kilele Archade (kilelearchade.co.ke) is that product: document-based learning, an AI tutor that cites the source, quizzes with explanations, and separate dashboards for learners, teachers and parents.
The architecture in one paragraph
React and TypeScript on the front. Supabase and PostgreSQL behind it — with pgvector holding embeddings next to the relational rows for subjects, levels and competences, so retrieval is a query, not a separate service. Curriculum documents flow through an ingestion pipeline into chunks and vectors; the tutor answers only from retrieved context, with a fallback provider layer so a single AI vendor outage never reaches a student.
What the classrooms changed
Building alone gets you to 'it works'. Real users get you to 'it works for them'. Early teacher feedback moved the product more than any technical decision: answers had to show which document page they came from before teachers would trust the tutor at all; the tutor may explain and hint, but assessment sessions run with that context locked away.
Solo, but not unmanaged
One engineer holding product, frontend, backend, AI pipeline and deployments forces boring choices: one database, one hosting story, scripts for anything done twice, and a retrieval bench (see the Lab) so configuration changes are measured instead of argued with myself.
Where it goes from here
Kilele Archade is live at kilelearchade.co.ke and still evolving through classroom feedback. The next fronts are broader subject coverage, offline-friendly access for areas with patchy connectivity, and better analytics so a teacher sees a struggling student before report day does.