About RecallShield

Inspiration

A product recall can put families at risk, yet many people never hear about it in time. We realized that safety information is only useful when it reaches the people who need it. While people may receive alerts for products purchased online, their homes also contain gifts, second-hand items, in-store purchases, vehicles, food, and medicines. This inspired us to build RecallShield, an AI-powered household safety assistant that helps families identify recalled products and receive timely, proactive warnings through a voice-enabled assistant.

What We Built

RecallShield connects a conversational voice assistant with official recall information from the U.S. Consumer Product Safety Commission (CPSC), National Highway Traffic Safety Administration (NHTSA), and FDA's openFDA data sources. Families can register household products by voice or text, check individual items or their entire inventory, review safety alerts, and learn about recommended remedies such as repairs, replacements, or refunds.

Built using the Model Context Protocol (MCP) and AWS serverless services, the system uses Amazon Bedrock for AI reasoning, Amazon Transcribe for speech recognition, Amazon Polly for spoken responses, AWS Lambda for backend processing, DynamoDB for household inventories and recall data, and EventBridge for scheduled recall monitoring. A simulated Alexa+ interface demonstrates how the assistant can communicate with the MCP server and proactively notify families when a new recall affects their belongings.

How We Built It

We developed a TypeScript-based MCP server with dedicated tools for registering products, checking recalls, managing household inventories, and resolving alerts. The backend retrieves official recall information and maintains a cached copy to improve resilience when external APIs are unavailable. A matching pipeline compares product brands, model numbers, production periods, and other identifying details. When information is incomplete, the assistant asks clarifying questions rather than making unsupported safety claims.

We also implemented a scheduled monitoring workflow that checks for new or updated recalls and matches them against registered household items. The application is deployed using AWS serverless infrastructure, with automated tests and security controls designed to make the system reliable and cost-conscious.

Challenges We Faced

One of our biggest challenges was matching real-world product descriptions to official recall records accurately. People may mispronounce brand names, omit model numbers, or describe products differently from government databases. We addressed this through data normalization, model-aware matching, clarification questions, and an AI-based second review that can downgrade uncertain matches but cannot upgrade them to confirmed recalls.

Another challenge was building a proactive alert system rather than a tool that only responds when someone asks a question. We designed scheduled monitoring and household-level matching to help identify newly announced recalls. Integrating speech services, AI reasoning, external government data, and serverless infrastructure also required careful testing, error handling, and security-conscious design.

What We Learned

Building RecallShield taught us that useful AI systems require more than a capable language model. Reliable data, deterministic validation, transparent uncertainty, automated testing, and responsible handling of safety-critical information are equally important. We gained practical experience with MCP, AWS serverless architecture, voice interfaces, recall-data integration, and automated evaluation.

Most importantly, we learned that technology can make existing safety information more actionable by delivering it through an interface people already know how to use.

Our Vision

With RecallShield, we want to make household safety proactive rather than reactive. Our vision is a future where families do not have to remember to search for recalls—their trusted assistant helps them stay informed, understand the risks, and take action when it matters.

RecallShield — Detect Recalls. Protect Lives.