{
  "id": 11225311,
  "title": "How uniopen customized Amazon Nova to their retail moderation policies for production deployment",
  "url": "https://urgent.news/2026/10/01/how-uniopen-customized-amazon-nova-to-their-retail-moderation",
  "topic": "ai",
  "section": "AI",
  "published": "2026-10-01T15:33:01.000Z",
  "source": {
    "name": "AWS Machine Learning",
    "slug": "aws-machine-learning",
    "url": "https://aws.amazon.com/blogs/machine-learning/how-uniopen-customized-amazon-nova-to-their-retail-moderation-policies-for-production-deployment/"
  },
  "original_language": "en",
  "account": "Uni-President Enterprises Group's digital communication and membership platform, uniopen, follows a moderation policy that classifies each interaction along two axes: behavior (nine categories) and subject (brand, other, or forbidden). To customize Amazon Nova 2 Lite to these specific policies, the team used Amazon SageMaker AI for supervised fine-tuning, followed by prompt-level output optimization. The AWS solution managed data, training, evaluation, and deployment in a single repeatable workflow. Amazon Nova 2 Lite processes primary moderation requests, while Amazon Nova 2 Pro generates candidate corrections for reported errors, which must be verified by a human reviewer before being added to the training set. Corrections are stored in Amazon S3 and tracked in Amazon DynamoDB, while orchestration, evaluation, and deployment are handled by Argo Workflows on Amazon EKS, Amazon SNS, and Amazon CloudWatch. To ensure model quality, hard and soft gates control promotion, with hard gates requiring regression tests and soft gates monitoring performance indicators. A conversation window treats a segment of a customer conversation as one training or evaluation example, and the model's performance is evaluated on a held-out test set of 737 conversation windows.",
  "summary": "See how uniopen, a retail platform from Taiwan's Uni-President Enterprises Group, adapted Amazon Nova 2 Lite to its content-moderation policies using supervised fine-tuning in Amazon SageMaker AI and prompt optimization. Business-relevant evaluation and release gates kept quality in check.",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 2,
    "also_reported_by": [
      {
        "outlet": "AWS Machine Learning",
        "title": "Build a multi-agent music production pipeline on Amazon Bedrock AgentCore Runtime Instances",
        "url": "https://urgent.news/2026/09/30/build-a-multi-agent-music-production-pipeline-on-amazon-bedrock",
        "published": "2026-09-30T15:21:57.000Z"
      }
    ]
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}