人工智能
Gen Digital 如何用 Amazon Bedrock Agent Core 网络搜索构建 Norton Revamp 的 AI 校验新闻引擎
你的职业名声现在先上网招募者在回电前检查LinkedIn. 客户在预约第一次会面前，先检查您的数字存在。 对于雄心勃勃的专业人士来说，这种转变是一个机遇：精心设计的在线声音可以打开一个简历永远无法打开的大门。
但站出来需要的不仅仅是内容生成。 这需要实时对当前情况的认识。 一个专业的建筑，作为一个基因化的AI思想领袖，需要react到今天的突破，而不是上周的头条新闻。 观众比以往更能奖励真实性和及时性。
AmazonBedrock Agent Core上的网络搜索使得这成为可能。 我们建立了Norton Revamp，一个基因化的AI动力专业声誉管理平台。 其核心是由AgentCore Web Search提供动力的AI精密新闻引擎，该引擎将每个内容建议，目标建议，以及当前所引用的网络知识的新闻素材作为依据。 2026年3月发布平台，由一小群工程师建造。
由Amazon Bedrock Agent Core Web Search协助客户将想法转化为社交媒体帖子
挑战：实时基于专业内容
与依赖静态知识基础的聊天员不同，诺顿Revamp需要了解目前每个用户专业领域的趋势。 网络安全专家需要今日的威胁情报。 市场领袖需要本周的产业转变。 传统方法(海关网络爬行器、第三方新闻API、静态内容数据库)都带来权衡：基础设施的间接费用、陈旧数据、供应商的复杂程度或数据流失风险。
在选择Core特工网络搜索之前我们评估了三种选择自定义爬行者需要持续维护并实时滞后。 第三方新闻API增加了供应商合同，API关键管理，数据离开我们的AWS环境。 静态知识库长期处于停滞状态。 代理Core Web搜索消除了这些担忧。 完全由零基础设施管理，完全在AWS范围内运行，零数据流转，这使得个性化，实时新闻素材在规模上可行。
代理 Core 网络搜索权限 Norton Revamp
Amazon Bedrock Agent Core上的Web搜索是一个完全管理起来的网络搜索能力，符合模式背景协议(MCP). 由Amazon Bedrock Agent Core提供。 我们的agent发送一个自然语言查询，Web搜索返回相关的片段，源 URL，标题，以及模型用来生成有底响应的发布日期。
这种能力具有三个核心特点：
AI 精密新闻种子
每个用户都会收到个性化内容流，如果你正在建立AI这个基因化思想领袖的声誉，您的饲料优先考虑AI突破，行业分析，与静态的RSS聚合器不同，feed是一款有目标意识的发现引擎，与每个会话一起刷新。
智能内容建议
用户接受AI生成的内容创意，适合自己空间中的趋势话题，用自己真实的声音书写，可选的AI生成视觉。 网络搜索探员对今天发生的事情，一个三步导师流(Analysis,Refine,Post)将用户从最初的想法行走到LinkedIn和X上发布内容，并带有反馈循环，因此AI在每一个迭代中会更加锐利。
趋势认识目标设定
用户定义一个目标。 AI绘制了个性化的路线图，其中包含每周的主题和里程碑。 AgentCore Web Search将实时趋势数据输入到目标生成中，因此每周的主题反映了当前的行业运动。 该系统从过去四周的活动中吸取经验教训，随着用户的增长而更新建议。
建筑和技术堆栈
整合遵循事件驱动架构。 当用户设定或更新其专业目标时，Amazon Nova(通过Amazon Bedrock)根据目标，最近的活动，以及接触模式生成定向搜索查询。 agent将这些查询发送给Agent Core Web Search，它返回了完全在我们AWS环境中引用的结果。 我们的AI学习系统在相关性、适切性和来源可信度方面对结果进行了排名。 排名结果随后成为Amazon Nova Lite在用户真实语音中生成内容建议的基础语境。 接触数据反馈到查询生成中，使新闻引擎逐渐变聪明。
基因AI和机器学习堆栈：
- Amazon Bedrock – 核心基因AI引擎用于内容生成，个性化，后分析(批量模式)，以及内容指导。
-Amazon Nova(英语：Nova Lite) – 主大型语言模型(英语：LLM)为内容生成，目标创建以及趋势分析提供动力。
- 在Amazon Bedrock Agent Core上进行网络搜索 – 实时网络检索为新闻引擎提供动力，内容定位，趋势发现通过MCP兼容的管理连接器。
-Stability AI – 图像生成。
-Claude(通过Amazon Bedrock) – 用于复杂推理任务的附加LLM能力。
基础设施：React,TypeScript，以及前端的Tailwind CSS. 节点Js,Amazon弹性集装箱服务(Amazon ECS),AWS Lambda,Amazon DynamoDB(带有复合密钥图案的多表),Amazon API Gateway,Amazon简易存储服务(Amazon S3),Amazon简易排队服务(Amazon SQS),Amazon EventBridge,Amazon CloudentFront,Amazon Cloudwatch，以及后端的AWS秘密管理器。 通过Terraform管理基础设施，在零下架时部署蓝绿色发布。
处理管道：该平台分为四个管道阶段：
- Full Scan – 拉上1000个历史哨所，存放于Amazon S3，运行Amazon Bedrock批量分析，以了解声誉影响。
- Daily Scan – Amazon EventBridge - 触发增量分析。
— Weekly Goal Generation – Lambda生成目标，考虑过去参与数据加上来自Agent Core Web Search的实时趋势数据。
- 内容教练 - 同步 Amazon 贝德罗克通过分析，完善，邮政流调用，每个步骤都以Agent Core Web搜索结果为基地。
具有AI-内在发展思维的工程
我们从第一天开始就嵌入基因AI，作为产品和发展方法。 人为-AI责任的分裂是有意的：人为的建筑决策、安全审查、业务逻辑和个人化算法。 AI处理代码生成，文档，测试脚手架，重构，以及UI组件创建。
有了Claude，我们每个概念产生了10个或更多的设计变体，压缩设计到原型的时间从数周到数小时，并且用AI生成了70%的UI代码。 我们在以前需要的时间内测试了五个定价模式。 游戏系统在发布前经历了8次迭代。
可测量的影响：3x更快的特性执行，锅炉的生成加速90%，文献记录80%，测试生成70%,UI组件生成65%。 权衡：人工智能产出审查消耗了我们20%至30%的时间。
AI帮助个人化算法，DynamomDB多表架构(需要基于经验的判断)，安全执行，以及Amazon Bedrock批处理优化。 主要见解：AI加速实施，但人类设计AI学习系统如何适应用户目标。 产品基因AI(Amazon Bedrock,Amazon Nova,Agent Core Web Search)和开发基因AI服务于不同的目的，我们将它们分开。
测试、安全和质量保证
我们实现了85%的代码覆盖通过自动化测试生成。 关键测试领域包括内容生成质量、AgentCore Web搜索结果相关性和新鲜度验证、接触跟踪准确性、游戏逻辑以及新闻反馈与用户目标的相关性。
对于带有诺顿品牌的产品来说，遵守是不可谈判的。 安全检查跨越静态代码分析，依赖性脆弱性扫描，认证流程，用户数据加密，访问控制逻辑，基因AI模型输出验证，以及内容过滤。 我们的安全小组审查了大赦国际提出的所有建议：大赦国际加快识别，人类作出最后决定。
结果：自发布与实时监测基因AI内容质量和用户满意度以来，恢复时间为99.9%。
代理Core网络搜索集成的最佳做法
对于考虑代理Core Web搜索实时地面的团队：
投资于查询生成 — — 结构良好的查询比开放式查询能产生更好的结果，所以我们专注于那里的快速工程。
策略性缓存 — 实时新鲜度是关键，但有些查询不需要重新执行每个会话。 我们以符合新鲜度要求的实时值(TTL)缓存结果。
- 建立反馈循环 – 跟踪网络搜索结果用户与哪些用户接触，并将该信号反馈到查询生成中。 我们的新闻引擎在头90天里更加实用。
- 验证源质量 – Agent Core Web Search 返回源URL和发布日期。 我们建立了轻量级信誉评分层，以优先确定权威来源。
结论
Building Norton Revamp证明，一个小型的，基因化的AI强化团队可以在几个月内交付企业软件。 这也证明了Amazon Bedrock Agent Core上的Web Search可以作为实时智能骨干，用于需要保持与世界同步的产品。
我们发布是一个基因化的AI本土专业声誉管理平台，数月内有个性化的目标设定，AI驱动的内容建议，实时编译的新闻素材基于AgentCore Web搜索，适应性学习，游戏化，以及诺顿级安全。 一小群工程师完成了99.9%的上线时间。
基因AI不会取代工程师。 它放大了它们。 与Agent Core Web Search等完全管理的服务对等放大，消除了整个类别的基础设施复杂性。 人类设计了视觉。 Generative AI通过Amazon Bedrock,Amazon Nova,Agent Core Web Search加速使其复活，基于实时，引用了网络知识。
“对Amazon Bedrock Agent Core的网络搜索消除了建设和维护我们自己的网络检索基础设施的复杂性。 它让我们的诺顿Revamp新闻引擎传送实时，我们几个月后从概念转向生产
为了更多地了解Amazon Bedrock Agent Core上的Web搜索，访问AgentCore文档。 看看诺顿Revamp。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
Artificial Intelligence
How Gen Digital built Norton Revamp’s AI-curated News Engine with Amazon Bedrock AgentCore Web Search
Your professional reputation now lives online first. Recruiters check LinkedIn before returning a call. Clients review your digital presence before booking a first meeting. For ambitious professionals, that shift is an opportunity: a well-crafted online voice can open doors that a résumé alone never could.
But standing out takes more than content generation. It takes real-time awareness of what’s happening right now. A professional building a reputation as a generative AI thought leader needs to react to today’s breakthroughs, not last week’s headlines. And with feeds flooded by AI-generated noise, audiences reward authenticity and timeliness more than ever.
Web Search on Amazon Bedrock AgentCore made that possible. We built Norton Revamp, a generative AI-powered professional reputation management platform. At its heart is an AI-curated News Engine, powered by AgentCore Web Search, that grounds every content suggestion, goal recommendation, and news feed item in current, cited web knowledge. The platform launched in March 2026, built by a small group of engineers.
In this post, we show you how the Gen team builds an AI-curated News Engine in Norton Revamp with Amazon Bedrock AgentCore Web Search to help customers turn ideas into engaging social media posts.
The challenge: Real-time grounding for professional content
Unlike chatbots that rely on static knowledge bases, Norton Revamp needs to understand what’s trending right now in each user’s professional domain. A cyber security expert needs today’s threat intelligence. A marketing leader needs this week’s industry shifts. Traditional approaches (custom web crawlers, third-party news APIs, static content databases) each carry trade-offs: infrastructure overhead, stale data, vendor complexity, or data egress risk.
We evaluated three alternatives before choosing AgentCore Web Search. Custom crawlers require ongoing maintenance and lag real time. Third-party news APIs added vendor contracts, API key management, and data leaving our AWS environment. Static knowledge bases were perpetually stale. AgentCore Web Search eliminated these concerns. It’s fully managed with zero infrastructure overhead, operates entirely within AWS with zero data egress, and returns cited sources with URLs and publication dates. This makes personalized, real-time news feeds viable at scale.
How AgentCore Web Search powers Norton Revamp
Web Search on Amazon Bedrock AgentCore is a fully managed web search capability, compliant with the Model Context Protocol (MCP). It’s available as a managed connector on the AgentCore Gateway, a capability of Amazon Bedrock AgentCore. Our agent sends a natural-language query, and Web Search returns relevant snippets, source URLs, titles, and publication dates that the model uses to produce grounded responses.
This capability powers three core features:
AI-curated news feed
Every user gets a personalized content stream surfacing the latest articles and trending topics tied to their goal. If you’re building a reputation as a generative AI thought leader, your feed prioritizes AI breakthroughs, industry analysis, and emerging use cases, sourced in real time through AgentCore Web Search. Unlike a static RSS aggregator, the feed is a goal-aware discovery engine that refreshes with every session.
Smart content suggestions
Users receive AI-generated content ideas tailored to trending topics in their space, written in their own authentic voice, with optional AI-generated visuals. AgentCore Web Search grounds every suggestion on what’s happening today. A three-step coaching flow (Analyze, Refine, Post) walks users from initial ideas to published content on LinkedIn and X, with a feedback loop, so the AI gets sharper with every iteration.
Trend-aware goal setting
Users define a goal. The AI builds a personalized roadmap with weekly themes and milestones. AgentCore Web Search feeds real-time trend data into goal generation, so weekly themes reflect current industry movements. The system learns from the past four weeks of activity, refreshing recommendations as users grow.
Architecture and tech stack
The integration follows an event-driven architecture. When a user sets or updates their professional goal, Amazon Nova (through Amazon Bedrock) generates targeted search queries based on the goal, recent activity, and engagement patterns. The agent sends these queries to AgentCore Web Search, which returns cited results entirely within our AWS environment. Our AI Learning System ranks the results for relevance, recency, and source credibility. The ranked results then serve as grounding context for Amazon Nova Lite to generate content suggestions in the user’s authentic voice. Engagement data feeds back into query generation, making the News Engine progressively smarter.
Generative AI and machine learning stack:
- Amazon Bedrock – Core generative AI engine for content generation, personalization, post analysis (batch mode), and content coaching.
- Amazon Nova (Nova Lite) – Primary large language model (LLM) powering content generation, goal creation, and trend analysis.
- Web Search on Amazon Bedrock AgentCore – Real-time web retrieval powering the News Engine, content grounding, and trend discovery through an MCP-compliant managed connector.
- Stability AI – Image generation.
- Claude (through Amazon Bedrock) – Additional LLM capabilities for complex reasoning tasks.
Infrastructure: React, TypeScript, and Tailwind CSS on the frontend. Node.js, Amazon Elastic Container Service (Amazon ECS), AWS Lambda, Amazon DynamoDB (multi-table with composite key patterns), Amazon API Gateway, Amazon Simple Storage Service (Amazon S3), Amazon Simple Queue Service (Amazon SQS), Amazon EventBridge, Amazon CloudFront, Amazon CloudWatch, and AWS Secrets Manager on the backend. Infrastructure managed through Terraform with blue-green deployment for zero-downtime releases.
Processing pipeline: The platform runs four pipeline stages:
- Full Scan – Pull up to 1,000 historical posts, store in Amazon S3, run Amazon Bedrock batch analysis for reputation impact.
- Daily Scan – Amazon EventBridge-triggered incremental analysis.
- Weekly Goal Generation – Lambda generates goals considering past engagement data plus real-time trend data from AgentCore Web Search.
- Content Coaching – Synchronous Amazon Bedrock calls through the Analyze, Refine, Post flow, each step grounded with AgentCore Web Search results.
Engineering with an AI-native development mindset
We embedded generative AI from day one, as both the product and the development methodology. The human-AI responsibility split was deliberate: human-owned architecture decisions, security reviews, business logic, and personalization algorithms. AI handled code generation, documentation, test scaffolding, refactoring, and UI component creation.
With Claude, we produced 10 or more design variants per concept, compressed design-to-prototype time from weeks to hours and generated 70 percent of UI code with AI. We tested five pricing models in the time previously needed for one. The gamification system went through eight iterations before launching.
The measurable impact: 3x faster feature implementation, boilerplate generation accelerated by 90 percent, documentation by 80 percent, test creation by 70 percent, and UI component generation by 65 percent. The trade-off: AI output review consumed 20–30 percent of our time.
AI helped with personalization algorithms, the DynamoDB multi-table architecture (which needed experience-based judgment), security implementation, and Amazon Bedrock batch processing optimization. The key insight: AI accelerated implementation, but humans designed how the AI Learning System should adapt to user goals. Product generative AI (Amazon Bedrock, Amazon Nova, AgentCore Web Search) and development generative AI served different purposes, and we kept them separate.
Testing, security, and quality assurance
We achieved 85 percent code coverage through automated test generation. Critical test areas included content generation quality, AgentCore Web Search result relevance and freshness validation, engagement tracking accuracy, gamification logic, and news feed relevance to user goals.
For a product carrying the Norton brand, compliance is non-negotiable. Security checks spanned static code analysis, dependency vulnerability scanning, authentication flows, user data encryption, access control logic, generative AI model output validation, and content filtering. Our security team reviewed all AI-generated recommendations: AI accelerated identification, and humans made final decisions.
The result: 99.9 percent uptime since launch with real-time monitoring of generative AI content quality and user satisfaction.
Best practices for AgentCore Web Search integration
For teams considering AgentCore Web Search for real-time grounding:
- Invest in query generation – Well-structured queries yield better results than open-ended ones, so we focused on our prompt engineering there.
- Cache strategically – Real-time freshness is the point, but some queries don’t need to re-execute every session. We cache results with time-to-live (TTL) values that match freshness requirements.
- Build feedback loops – Track which web search results users engage with and feed that signal back into query generation. Our News Engine’s relevance improved over the first 90 days.
- Validate source quality – AgentCore Web Search returns source URLs and publication dates. We built a lightweight credibility scoring layer to prioritize authoritative sources.
Conclusion
Building Norton Revamp proved that a small, generative AI-augmented team can deliver enterprise software in months. It also proved that Web Search on Amazon Bedrock AgentCore can serve as the real-time intelligence backbone for products that need to stay current with the world.
We launched a generative AI-native professional reputation management platform in months, with personalized goal setting, AI-driven content suggestions, real-time curated news feeds grounded by AgentCore Web Search, adaptive learning, gamification, and Norton-grade security. A small group of engineers achieved 99.9 percent uptime.
Generative AI doesn’t replace engineers. It amplifies them. Pairing that amplification with fully managed services like AgentCore Web Search removes entire categories of infrastructure complexity. Humans designed the vision. Generative AI through Amazon Bedrock, Amazon Nova, and AgentCore Web Search accelerated bringing it to life, grounded in real-time, cited web knowledge.
“Web Search on Amazon Bedrock AgentCore eliminates the complexity of building and maintaining our own web retrieval infrastructure. It lets our Norton Revamp News Engine deliver real-time, personalized content grounded in what’s happening in each user’s professional domain, with zero data egress and Norton-grade privacy. We went from concept to production in months, and AgentCore Web Search was a key enabler.”
To learn more about Web Search on Amazon Bedrock AgentCore, visit the AgentCore documentation. Check out Norton Revamp.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.