curio, an AI-powered news brief

Autonomous AWS agent that fetches, analyzes, and publishes verified news digests, saving time and reducing doomscrolling

Comment 3

Inspiration:

I love to start my commute to work by listening to a 5-minute-long news-brief podcast to stay informed with humor and a comfortable, yet edgy tone. However, I only like one from Spain. I aim to develop an autonomous news agent that delivers concise, reliable updates, freeing people from the habit of doomscrolling while still keeping them informed.

What it does

This Flash News media generator is an AI agent on AWS that: Fetches news from trusted sources. Uses reasoning and RAG to fact-check and prioritize relevance. Summarizes into short, personalized digests. Outputs in multiple formats: single-page web app, audio brief (via Amazon Polly), and short-form video scripts. Publishes automatically to email, web, and social platforms.

How we built it

Amazon Bedrock (Claude/Nova): LLM reasoning and summarization. Bedrock AgentCore: orchestration of tool calls (fetch, fact-check, publish). Lambda + EventBridge: scheduled ingestion of sources into S3. Knowledge Bases + DynamoDB: grounding and storing curated articles. Step Functions: workflow automation from ingest → verify → summarize → publish. Amazon Polly + SES + CloudFront: audio digest generation, email distribution, and static site delivery.

Challenges we ran into

Making the agent feel autonomous rather than scripted. Ensuring summaries were concise yet accurate. Orchestrating multiple AWS services in a short hackathon timeline. Managing fact-checking with limited time and data sources.

Accomplishments that we're proud of:

Built a working end-to-end pipeline that runs daily and produces news digests automatically. Created multi-modal outputs (text, audio, video-ready scripts). Showcased a practical use case of Bedrock AgentCore + Knowledge Bases for real-world problems.

What we learned

How to implement tool schemas and planning with Bedrock AgentCore. Best practices for chaining AWS services with Step Functions. The importance of balancing automation vs. human trust in news delivery. How AWS makes it possible to build a robust agent with relatively lean code.

What's next for flash news media generator

Add user personalization: choose topics, tone, and length of digest. Expand into local news and niche briefings (e.g., startups, AI, sports). Integrate with Amazon Q for interactive Q&A on “what did I miss?” Explore sponsorships and affiliate models to sustain the service. Scale the architecture for global, multi-language coverage. Including a chatbot for follow-up questions, so it feels more interactive.

All designed and deployed using Kiro!

Built With

  • amazon-bedrock
  • amazon-polly
  • amazon-q
  • amazon-web-services
  • aws-lambda
  • bedrock-agentcore
  • cloudfront
  • dynamodb
  • eventbridge
  • ses
  • step-functions
Share this project:

Updates