Workflow · Production

Duolingo builds BirdBrain ML model to personalize lesson difficulty for each learner

The problem

Without adaptive difficulty, learners either wasted time repeating content they already knew or struggled with material beyond their current level, making it impossible to teach the right thing at the right time.

Workflow diagram · grounded in source
1
Learner completes exercises
trigger
“バードブレインは、Duolingoの学習者が練習問題を多くこなせばこなすほど、学習者の習熟度や、練習問題がどの程度難しかったかを学習していきます”
2
BirdBrain predicts correctness
ai_action
“バードブレインはこれらの情報を組み合わせ、学習者が出題された練習問題に正解できるかどうかを、データに基づき推測します”
3
Session Generator selects questions
routing
“セッション・ジェネレーターは、バードブレインが開発されたおかげで、大量の練習問題の候補の中から、学習者に最適な難易度の問題を選び出すのに必要な情報を得ることができるようになりました”
4
Personalized lesson delivered
output
“バードブレインは、すでに学習者が無理なく解答できるトピックは自動的に避けてくれるため、やりがいのある問題に集中することができます”
5
Model learns from all learner data
feedback_loop
“バードブレインは、学習者の学習データから得られた情報を言語学習のあらゆる側面に活かし、アプリをより優れたものとする役目を果たしています”
Reported outcome

A/B testing showed BirdBrain-powered lessons consistently improved learning outcomes; learners tackled more problems and continued daily learning more often, and the overall learning experience was significantly improved.

Reported metrics
Learning outcomes improvement一貫して学習効果が上がる (consistently improved learning outcomes)
Daily learning continuation毎日コツコツと学習を継続する傾向 (tendency to continue learning daily)
Learning experience improvement学習体験が大幅に改善された (learning experience significantly improved)
Reported stack
バードブレインセッション・ジェネレーター
Source
https://blog.duolingo.com/ja/duolingo-ai-model/
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

A/B testing showed BirdBrain-powered lessons consistently improved learning outcomes; learners tackled more problems and continued daily learning more often, and the overall learning experience was significantly impro…

What tools did this team use?

バードブレイン, セッション・ジェネレーター.

What results were reported?

Learning outcomes improvement: 一貫して学習効果が上がる (consistently improved learning outcomes); Daily learning continuation: 毎日コツコツと学習を継続する傾向 (tendency to continue learning daily); Learning experience improvement: 学習体験が大幅に改善された (learning experience significantly improved) (source-reported, not independently verified).

How is this workflow AI workflow structured?

Learner completes exercises → BirdBrain predicts correctness → Session Generator selects questions → Personalized lesson delivered → Model learns from all learner data.