MedGuide Agent

An autonomous healthcare safety agent that detects harmful drug combinations, educates patients, and alerts doctors, powered entirely by AWS AI stack.

Comment

Inspiration

Every single day, patients across the world are harmed by the very medications meant to heal them. The World Health Organization reports that 1 in 30 patients experience adverse drug reactions over 260 million lives lost each year. Medication errors lead to millions of hospital admissions annually and cost over $42 billion globally, three-quarters of which are completely preventable. This isn’t just a clinical challenge; it’s a global systems failure. MedGuide AI was born to fix that by giving every clinician and patient a real-time, reasoning-powered safety assistant.

What it does

MedGuide AI is an autonomous clinical safety agent that detects unsafe medication combinations, explains the reasoning, and generates a QR-verified clinical report within seconds. Users simply enter gender, age, health conditions, allergies, and medications manually, by category, or via quick-add.

The agent performs:

  • Real-time drug safety checks via OpenFDA API
  • LLM-powered reasoning through Amazon Bedrock
  • Secure report generation with AWS Lambda and S3

Patients and doctors can view verified results instantly or scan a QR code to access the report from any device. It’s mobile-friendly, private, and designed for universal accessibility from hospitals to small rural clinics.

How we built it

  • Frontend: HTML + Tailwind + JavaScript for the web app
  • Backend: Node.js + AWS Lambda + API Gateway for serverless logic
  • AI Core: Amazon Bedrock (LLM) for reasoning and explainability
  • Data Source: OpenFDA API for verified medication and interaction data
  • Storage & Security: Amazon S3, IAM, and KMS encryption
  • Analytics: CloudWatch for performance logs and monitoring
  • The system runs fully serverless, ensuring speed, scalability, and HIPAA-aligned security.

Challenges we ran into

  • Fine-tuning prompt reasoning to ensure clinically relevant outputs
  • Integrating OpenFDA with Bedrock in real-time for live reasoning
  • Optimizing response latency while maintaining accuracy under 10 seconds
  • Designing a user experience intuitive for both clinicians and patients

Each challenge pushed us to make the system more stable, transparent, and secure.

Accomplishments that we're proud of

  • Built the first fully autonomous medication safety agent on AWS
  • Generated instant QR-verifiable safety reports for true transparency
  • Designed a mobile-first interface for universal accessibility
  • Achieved analysis + reasoning + report generation under 10 seconds
  • Created a serverless AWS architecture validated end-to-end

No direct competitors:

  • Existing tools like Lexicomp and Micromedex are static, paid databases.
  • IBM Watson Health was discontinued in 2022.
  • Google MedLM and Med-PaLM focus on Q&A, not verifiable reporting.
  • Epic EHR CDS modules are region-locked and unaffordable for smaller clinics.

MedGuide AI stands alone as the first-to-market autonomous agent that can reason, verify, and report medication safety in real time, creating a new category rather than competing in one.

What we learned

  • Deep integration between LLM reasoning and real-world APIs requires careful orchestration for reliability and compliance.
  • AWS’s serverless stack (Lambda + Bedrock + S3) enables healthcare-grade scalability.
  • Simplicity in UX is as powerful as sophistication in AI, especially for patient trust.

What's next for MedGuide Agent

  • Launching pilot testing across urban and rural clinics in Sri Lanka (2025)
  • Expanding datasets to include global drug registries (EMA, WHO, DrugBank)
  • Adding multilingual support for English, Sinhala, and Tamil
  • Developing clinician dashboards and integration with EHR systems
  • Moving toward certification for FDA Software as a Medical Device (SaMD)

MedGuide AI’s long-term vision is to make medication safety autonomous, explainable, and accessible everywhere.

Built With

Share this project:

Updates