How Duolingo Cut Translation Costs by 40 Percent with a Custom MT Pipeline

Episode notes

Lucas and Luna dig into how Duolingo slashed translation costs by 40 percent by building a custom machine-translation pipeline for its course content. They walk through the tradeoffs between generic MT engines and a tailored model, how the team balanced quality with speed, and what lessons any engineering org can take when considering build-versus-buy for AI. Along the way, they touch on the surprising role of human reviewers and the economics of running your own inference infrastructure. A concrete case study for anyone wrestling with AI costs and quality at scale. #Duolingo #MachineTranslation #CustomMT #CostReduction #AIInfrastructure #EngineeringOrg #BuildVsBuy #NLP #BusinessAndTechnology #Podcast #FexingoBusiness #BusinessPodcast #TechLeadership #CTO #TranslationPipeline #InferenceCosts #QualityAssurance #HumanReview Keep every episode free: buymeacoffee.com/fexingo

Related episodes across the Index

Other episodes covering the same guests and topics, from across The B2B Podcast Index.

• Will GenAI Replace Developers? with Nik FroehlichSaaS Scaled · on build vs buy

• How Datadog Scaled Engineering Without Burning Out82 / 100
• How Slack Scaled Engineering Without Adding Headcount
• How Atlassian Tamed Technical Debt With Architecture Contracts
• How Atlassian Tamed Technical Debt With Architecture Contracts
• How GitHub Tamed Merge Conflict Chaos