Project Story — Flamingo AI

Inspiration

In a crowded market of language apps, most products follow the same script: fixed lesson flows, repetitive exercises, and one-size-fits-all content. Learners get bored, progress stalls, and cultural nuance is ignored. We wanted an app that feels alive — one that adapts to a learner’s level, cultural background, goals, and mood. Flamingo AI was inspired by that gap: a desire to make language learning adaptive, customizable, and genuinely joyful.

What it does

Flamingo AI is an AI-driven, offline-friendly Progressive Web App that delivers personalized language lessons and practice exercises.

Core features:

AI-Generated Lessons: Dynamic exercises (multiple-choice, short answer, contextual dialogues) generated on demand based on user level and interests.
Adaptive Difficulty: Content evolves with the learner — the AI adjusts difficulty, vocabulary, and scenarios.
Multilingual & Cultural Context: Support for arbitrary language pairs and culturally appropriate examples.
Gamification & Motivation: XP, streaks, and progress tracking to keep learners engaged.
Offline-First Experience: Fully usable without network connectivity; syncs progress when online.
Accessibility & Localization: Screen-reader support, keyboard navigation, and easily translatable UI.

How we built it

We chose a modern, testable stack focused on performance and developer velocity.

Frontend & tooling

React 19 + TypeScript 5.8
Tailwind CSS for styling
Vite 7, Workbox service worker, code splitting and lazy loading for PWA behavior

AI & orchestration

Kiro for orchestrating AI workflows and calling external LLMs for content generation and automated QA

Testing & deployment

Vitest + Playwright for unit & E2E tests
Automated accessibility checks and content validation scripts Framing progression this way guided how we design content generation and difficulty ramps.

Challenges we ran into

Balancing AI creativity vs. curriculum coherence. Free-form generation is powerful but can produce inconsistent difficulty or bad examples; we added validation layers and templates to constrain outputs.
Offline capability. Shipping an offline-first experience meant careful caching strategies, compact model fallbacks, and robust sync logic.
Hackathon constraints. Timeboxed development forced prioritization: ship a highly polished core experience rather than many half-finished features.

Accomplishments we're proud of

A production-ready PWA with high Lighthouse scores and real offline usability.
A scalable multilingual framework where new language pairs can be added quickly and reliably.
Automated content validation and a testing suite covering UI, E2E flows, and content correctness.
Clean, accessible UI and a gamified experience that keeps learners returning.

What we learned

Orchestrating AI workflows (via Kiro) requires both creative prompts and guardrails — prompts alone aren’t enough.
Cultural context materially improves retention and engagement; literal translations aren’t enough.
Offline-first apps demand a different mental model for sync, state reconciliation, and UX.
Good engineering practices (types, tests, CI) pay off hugely when iterating fast under pressure.