Popsa uses Amazon Nova and Claude 3 Haiku via Amazon Bedrock to generate personalized photo book titles across 12 languages, increasing positive user feedback by 13%.

Popsa reimagined its Title Suggestion feature—launched in 2021 using rule-based templates—by applying retrieval-based few-shot prompting (a technique where an LLM learns from example title-subtitle pairs similar to the user's photos) with Anthropic's Claude 3 Haiku and Amazon Bedrock.

The new approach automatically generates creative titles and subtitles that comply with strict requirements: each must stay under 36 characters, include a valid category (to determine which icon displays), and output valid JSON. Results improved across all metrics: positive user feedback rose by 13% (from 58% to 71%), and Design Created and Purchase metrics also improved.

In early 2025, multivariate A/B testing showed that Amazon Nova Pro models achieved higher user satisfaction than Claude 3 Haiku (71% positive feedback). The system generated over 5.5 million personalised titles in 2025, operating across more than 50 countries.