Adekunle Omorotimi

AI adaptive learning · Sumeru Tech

SchoolExl / TrainExl

An AI learning platform with voice conversations and streaming tutoring, live with about 1,000 learners in India and Jamaica. I architected the frontend, the layer that connects it to language models, and a design system that four products now share.

role
Software Engineer Lead · frontend and LLM-integration architecture
organisation
Sumeru Tech
period
Jun 2025 – present
  • ~1,000

    learners live across India and Jamaica

    source: CV, Sumeru Tech

  • 4

    products reusing the design system

    source: CV (Full-Stack & AI), Sumeru Tech

The problem

In an adaptive learning product, the model's answer is only half the work; the rest is how it becomes an interface. Responses stream in, voice has to take turns, and state has to stay coherent while a learner moves between lessons, assessments and the school portal, in more than one language.

What I built

I architected the React and Next.js (TypeScript) frontend and the integration layer between it and the language models: streaming responses, conversation state, and interfaces that change with what the learner is doing.

Voice-based learning, including multilingual voice conversations, runs on the same layer, alongside assessment workflows and the school portal.

I worked closely with the AI and backend engineers on data flow, response handling and model-to-UI communication, maintained the CI/CD workflow, and set frontend standards and design consistency across the suite.

The component work became a design system that is now reused across four products.

SchoolExl launched publicly in October 2025, an initiative of The Art of Living and Sumeru Edu Tech. The launch coverage states an aim of reaching more than 100,000 children across 1,327+ free schools in 22 states; the ~1,000 learners below is the live deployment I can vouch for.

From learner to model and backHow one turn moves through the platform.
person / result

Learner. Types, speaks or answers an assessment item.

Diagrams are simplified from the work described above. Node colours: model, code, external service, person or result.

Screens

  • ScoreExl Mentors Hub: a searchable grid of AI subject coaches such as Concept Coach, Vocab Coach, SEL Coach and Exam Prep Coach, each with a details button.
    Mentors Hub (design)The Mentors Hub design: a searchable library of AI subject coaches (Concept, Vocab, SEL, SDG, STEM, Exam Prep, Startup), each with details and a save action, alongside a personal 'My Hub'.
  • ScoreExl student home with search, voice input and four modules: Learning Centre, Mentors Hub, Exam Preparation and AI Insights.
    Student homeThe student home: search with voice input, and four modules (Learning Centre, Mentors Hub, Exam Preparation, AI Insights).
  • An AI-generated, colour-coded concept mind map for a topic.
    Concept mind mapAn AI breaks a topic down and the frontend lays it out as an interactive, colour-coded mind map: structured knowledge as data visualisation.
  • The teaching co-pilot with a concept mind map and feature tabs for presentations, quizzes and flash cards.
    Teaching co-pilotThe map lives inside a co-pilot that also generates presentations, quizzes, flash cards and activities: different ways to render the same content.
  • An interactive simulation where inputs drive a live animated model of a concept.
    Interactive simulationAn interactive simulation where inputs drive a live, animated model of the concept.
  • A data-rich concept whiteboard inside the AI teaching co-pilot.
    Visual explainerA data-rich concept whiteboard inside the AI teaching co-pilot.
  • Learner analytics dashboard with a performance line, subject-score bars, an activity donut and a deadline table.
    AI InsightsLearner analytics: performance over time, subject scores, an activity donut and a deadline table with status chips.
  • The same screen in an ocean colour theme.
    Theme: OceanThe UI runs on CSS design tokens, so one screen re-skins instantly. Ocean theme.
  • The same screen in a mint colour theme.
    Theme: MintMint theme: same screen, same components.
  • The same screen in a lavender colour theme.
    Theme: LavenderLavender theme: the token system carries colours, gradients and accents.

Engineering decisions

  1. 01

    One integration layer, several surfaces

    Chat, voice and assessment screens reach the models through a single layer, so streaming, conversation state and error handling are solved once instead of per screen.

  2. 02

    A shared design system

    Reusing one component set across four products keeps behaviour consistent (focus, loading, errors) and lets a new product start from finished primitives.

Stack

Frontend
  • React
  • Next.js
  • TypeScript
AI
  • LLM APIs (OpenAI, Anthropic)
  • Streaming responses
  • Voice AI
  • Conversation state
Delivery
  • CI/CD
  • Design system

next case study

HODL
Discuss this work