Calm is a San Francisco–based health and wellness company, best known for its leading meditation and sleep app. Its platform helps millions improve their mental health, sleep, and overall wellbeing.
Challenge
Calm wanted to take its personalised recommendations to the next level, delivering more relevant and engaging content for users. While Amazon Personalize already powered recommendations, the team sought to add deeper contextual awareness to improve accuracy and user satisfaction.
The goal was to harness user engagement data and detailed content metadata to create recommendations that not only matched preferences but also felt personal, transparent, and aligned with each user’s mood and persona.
Solution
Firemind built a recommendation engine that combined a vector database, large language models (LLMs), and a hybrid search approach. This architecture enriched Calm’s recommendation process by generating detailed, contextual content descriptions using LLMs.
The solution also introduced explainable AI for recommendations, where each suggestion included a tailored justification highlighting how it suited the user’s needs – enhancing transparency, trust, and engagement.
Services used
- Amazon Bedrock
- AWS Lambda
- Amazon DynamoDB
- Amazon OpenSearch
The Results
- 100% of recommendations now have explainability
- 25% improvement in recommendation accuracy
- Added 5 new metadata properties for greater personalisation
- Increased contextual awareness in recommendations
- Enhanced user trust and engagement
