Case Study · 2024

Selve

AI English Speech Trainer for Sri Lankan learners

B2C Mobile AppGen AIEdTech

RoleHead of Product & Operations

MarketSri Lanka

ModelCSR Initiative → Freemium

StackFlutter · Amazon Bedrock

The Problem

Fear of Speaking — A Market Hiding in Plain Sight

User Pain PointBeginner English speakers avoid public practice due to social stigma — a private AI solution removes that barrier entirely.

Business ObjectiveLaunched as a CSR initiative targeting high adoption; long-term roadmap to freemium model for operational sustainability.

Target AudienceSmartphone users below B1 English proficiency from middle and lower socioeconomic classes using mid- to low-end devices.

Research & Discovery

Insight-Led from Day One

MethodsInterviews, surveys, and continuous focus group discussions during both POC and MVP stages.

Competitor AnalysisAssessed offline learning, remote classes, AI-powered apps, and traditional tutoring — each had a critical gap in conversational feedback.

Market InsightAdvances in Gen AI significantly enhanced speech recognition, enabling natural conversations with detailed, personalised feedback at scale.

Design & UX

Accessible by Design

Mobile-FirstDesigned exclusively for smartphones — the primary (and often only) connected device for the target audience.

Bilingual UISimple bilingual interface accommodating low English literacy levels while encouraging progression toward full English interaction.

LightweightOptimised for mid- to low-end devices — performance on a $100 Android was a first-class requirement, not an afterthought.

UI Showcase

Core Features

What Makes Selve Different

Gamification

Leaderboards, storylines, points, and rewards designed to build a daily speaking habit.

AI Conversation

Custom RAG models for contextual conversation practice with real-time pronunciation feedback.

Social Elements

Interactive community walls and peer challenges to reduce isolation in the learning journey.

Adaptive UX

Intuitive dashboard and simplified learning paths that adjust to each learner's proficiency level.

Delivery

How We Built It

VisionTo become a trusted daily companion for users to practice and enhance English speaking skills without judgment.

PrioritisationUser research translated into RICE-scored feature lists, ensuring we shipped what moved retention — not what was easiest to build.

ProcessCollaborated with an outsourced dev team through POC → MVP → Final Product, managing scope, milestones, and wireframes throughout.

Learnings

Challenges We Solved

Language Barriers

Made the app bilingual, but encouraged switching to English at advanced levels for better learning outcomes.

API Latency

Optimised transcription, AI response, and text-to-voice pipelines to deliver smooth, uninterrupted conversation flow.

Cost Control

Balanced performance and inference costs using optimised Amazon Bedrock configurations without degrading quality.

Metrics

What We Measure

User MetricsDownloads, DAU/MAU stickiness, average session length, churn rate, and time between practice sessions.

PerformanceApp launch time, API response time, and AI transcription accuracy across device tiers.

LearningUser progress through evaluation scores and qualitative feedback from learners at each proficiency stage.

RevenueARPU and time to first purchase as the freemium model matures.