Popsa: Using Amazon Bedrock + Amazon Nova for personalized title suggestions (Photo Books)

Use case typeContent generationUpdated Apr 27, 2026

Popsa uses Amazon Bedrock and the Amazon Nova family of models to reimagine its Title Suggestion feature for photo books. The feature combines metadata from customer photos, on-device computer vision, retrieval-based few-shot prompting, and model comparisons to generate creative, brand-aligned titles and subtitles across 12 languages. The solution enforces strict formatting and layout constraints, including valid JSON output and a 36-character limit for both title and subtitle.

Organization
Popsa
Industry
Consumer & Food
Location
United Kingdom
Published
April 2026

Reported outcomes

+13%

positive user feedbackOther quantified impact

−35%time to first suggestion1.4 secondstime to first suggestion (seconds)0.9 secondstime to first suggestion (seconds)5,500,000 titlespersonalized titles generated

Strategic outcomes

Customer experience & trustHigher customer satisfactionCustomer experience & trustImproved engagement and purchase ratesScale & capacityRolled out to all users

Primary read

Use case focus

Showing 2 of 2

  • 1Content generation
  • 2Prompt optimization
  • Improve the Photo Books Title Suggestion feature beyond rule- and graph-based suggestions.
  • Generate more creative titles and subtitles while enforcing strict layout and JSON formatting constraints.
  • Improve quality, response time, and cost across 12 supported languages.
  • Built retrieval-based few-shot prompting using a database of example photo books and accepted title suggestions.
  • Used Amazon Bedrock to compare models and run Claude 3 Haiku and Amazon Nova Lite, Micro, and Pro through a unified API.
  • Added Amazon Bedrock ConverseStream to stream tokens and return the first validated suggestion faster.
  • Combined metadata, on-device computer vision, reverse geocoding, and subject classification to create richer prompts.
Technologies
Amazon BedrockAmazon NovaAmazon Nova LiteAmazon Nova ProAmazon Nova MicroAmazon Bedrock ConverseStreamClaude 3 Haiku
  • Positive user feedback increased by 13%, from 58% to 71%.
  • Amazon Nova Pro achieved 73% positive feedback with the lowest negative feedback at 12%.
  • Average time to first suggestion fell from 1.41 seconds to 0.92 seconds, a 35% improvement.
  • The feature generated over 5.5 million personalised titles in 2025.
  • The rollout expanded to 100% of users.
Architecture

The Title Suggestion Service decrypts and processes each design, extracts timestamps, reverse-geocodes latitude/longitude data, classifies the subject of the design, and then passes a generated description into a retrieval-based few-shot prompting component. The system uses Amazon Bedrock's unified API with Claude 3 Haiku and Amazon Nova models, and streams output with ConverseStream so the client can display the first valid title-subtitle-category triplet immediately while additional suggestions continue in the background.

Sources & evidence1
Groundedness: 5/5Type: Blog PostPublished: Apr 27, 2026Publisher: AWS Machine Learning BlogEvidence: VendorConfidence: High
Primary source

AI-generated summary. Verify important details with the linked sources before relying on this case.

Explore related AI use cases
Customer evidence
More from Popsa
Provider evidence
AWS provider pageAWS ranked cases
Market context
Consumer & Food AI use casesConsumer & Food industry insightsUnited Kingdom country insights
AI patterns
RAGVision
Browse the catalog
All AI use casesAI use-case typesContent generation use cases

Was this useful?

Similar cases