MarketPulse AI

MarketPulse AI is a GenAI-powered hyperlocal campaign generator designed for global CPG brands to create personalized, culturally relevant marketing content across regions.

Comment

Inspiration

The inspiration for MarketPulse AI came from observing how fragmented and time-consuming modern marketing workflows have become. Marketing teams often juggle multiple tools for content creation, market research, and campaign management, leading to inefficiencies and inconsistent messaging.

I envisioned a unified AI-powered platform where intelligent agents could collaborate seamlessly - one agent researching market trends while another generates compelling content, all orchestrated through an intuitive interface. The recent advances in Amazon's Nova models and the emergence of Agent-to-Agent (A2A) protocols presented the perfect opportunity to build this vision into reality.

The goal was to create not just another AI tool, but an intelligent ecosystem where specialized agents work together like a high-performing marketing team, each bringing their expertise to deliver comprehensive marketing solutions.

What it does

MarketPulse AI is a multi-agent marketing automation platform built on AWS Bedrock AgentCore and Strands. It:

  • Takes a marketing goal (e.g., “Smart Fitness app targeting fitness enthusiasts under age 40 in New York”).
  • Uses an Orchestrator agent (running in a Streamlit UI on ECS Fargate) to coordinate tasks.
  • Calls three specialized agents via A2A: -- city_info_agent → fetches weather and local events. -- market_research_agent → provides market and audience insights. -- content_generation_agent → creates on-brand copy, image prompts/assets, and short video scripts.

Stores context and artifacts in AgentCore Memory and logs all activity to AgentCore Observability. Delivers outputs (copy, images, video scripts) back to the UI for review and download.

In short: It turns a high-level marketing idea into a data-informed, brand-safe, multi-channel campaign—with creatives and insights—using agent collaboration over A2A and centralized memory/observability.

How we built it

Architecture Design

I started with a modular architecture approach, designing four independent but interconnected systems:

  • City Information System - Weather and events data with A2A server capabilities
  • Market Research System - Web research and analysis using Tavily API
  • Content Generation System - Multi-modal content creation with Nova models
  • Streamlit Web Interface - Unified dashboard orchestrating all systems

Technology Stack Selection

  • Strands Agents SDK: Chosen for its elegant agent orchestration capabilities and built-in A2A support
  • Amazon Bedrock & Nova Models: Selected for cutting-edge multi-modal AI capabilities
  • Streamlit: Picked for rapid UI development with advanced features like real-time updates
  • AWS Services: Leveraged S3 for storage, AgentCore for deployment, and IAM for security

Challenges we ran into

  • Video Generation Timeouts Nova Reels video generation takes 10-20 minutes, causing timeout issues. Solution Built job monitoring system with persistent state tracking and resume capabilities.
  • A2A Protocol Integration Coordinating multiple agent systems while maintaining individual system integrity. Solution: Implemented proper service discovery and graceful fallback mechanisms.
  • State Management Across Systems Maintaining campaign state across multiple independent agent systems. Solution: Designed centralized campaign management with distributed execution.
  • Complex Workflow Simplification Making powerful multi-agent capabilities accessible to non-technical users. Solution: Designed intuitive step-by-step interfaces with smart defaults and helpful guidance.
  • Real-Time Feedback Providing meaningful progress updates for complex AI operations. Solution: Built detailed progress tracking with estimated completion times and intermediate results.
  • Streamlit deployment issue with AppRunner Streamlit relies on WebSockets and is not compatible with the ALB option provided by AWS AppRunner service. Solution: Deployed Streamlit UI as AWS ECS Fargate container and configured stickiness at ALB level.

Accomplishments that we're proud of

  • Production-Ready Architecture: Built a scalable, enterprise-grade multi-agent system
  • Seamless Integration: Successfully integrated 4 different AI systems with A2A protocol
  • User-Centric Design: Created an intuitive interface that makes complex AI accessible
  • Comprehensive Testing: Developed full test suites ensuring system reliability
  • Documentation Excellence: Provided detailed documentation for easy adoption and contribution

What we learned

  • Bedrock Agentcore components and capabilities.
  • Strands Agents features
  • A2A protocol
  • Developing UI with Streamlit
  • Deployment using AWS ECS service
  • Spec-driven development with Kiro
  • Vibe coding with Amazon Q and Kiro

What's next for MarketPulse AI

  • Implement Long-term memory features on the Orchestrator agent
  • Transform the Orchestrator agent into a role like Campaign Director.
  • Enhance the UI and protect it behind a sign-in form.
  • Add functionalities for Brand and Avatar management and consistency
  • Improve the response time of the overall campaign workflow.
  • Integrate Bedrock Agentcore Gateway and Identity features.
  • Improve observability.

Built With

  • a2a
  • agentcore
  • bedrock
  • ecs
  • fargate
  • nova
  • strands
  • streamlit
Share this project:

Updates