Best of both worlds - How we built an AI-aided content creation tool for language learning

Built an AI-aided content creation tool focused on language learning with a human-in-loop approach, balancing automation with human expertise

Used prompt engineering and LangChain for implementation, with GPT models as the core LLM backend, deployed on AWS

Achieved 85%+ acceptance rate from content editors by providing high-quality examples and using existing Babbel content for context, which helped reduce hallucinations

Key workflow steps include prompt generation, content creation, evaluation, and human review/editing with feedback loops

Started with low-risk, high-value use cases to build trust within the organization and demonstrate value to skeptical teams

Emphasized maintaining cultural relevance and educational quality standards while scaling content creation

Integrated evaluation criteria like inclusion, diversity and content quality metrics into the automated assessment

Uses modular architecture allowing for easy LLM swapping and integration with existing content management systems

Focuses on personalization through user interests and preferences rather than pure individual content generation

Faces ongoing challenges around content localization, quality evaluation at scale, and balancing automation with human expertise

Leverages existing Babbel content data to ground the AI outputs and maintain consistent teaching standards