CareCostAI

AI care team for transparent healthcare costs

Comment

🩺 Inspiration

Healthcare costs in the U.S. are notoriously confusing, inconsistent, and opaque. Even with public CMS data available, patients have no intuitive way to understand what a procedure will actually cost, or why prices vary so dramatically between hospitals.

We were inspired to create CareCostAI after hearing real patient stories — people delaying essential procedures because they couldn’t predict their out-of-pocket expenses. We wanted to bring clarity, fairness, and empathy to healthcare pricing through AI.

💡 What it does

CareCostAI is an AI-powered, multi-agent healthcare assistant that helps users understand and compare the cost of medical procedures across hospitals — in plain language.

When a user asks something like:

“Compare knee replacement costs in Boston,”

CareCostAI simulates an AI care team of seven specialists (nurse, medical coder, billing analyst, insurance officer, etc.) who collaborate live to interpret the query, analyze cost data, and generate a personalized, transparent report — streamed in real time.

It transforms complex CMS and hospital quality datasets into human-friendly insights, showing:

  • Average and range of hospital costs
  • Insurance-adjusted, out-of-pocket estimates
  • Hospital recommendations based on affordability and quality
  • Easy-to-understand summaries using Gemini 1.5 Flash

⚙️ How we built it

We built CareCostAI as a serverless, multi-agent system on Google Cloud Run.

Architecture highlights:

  • Frontend: Next.js + TailwindCSS for live agent visualization and user interaction
  • Backend: FastAPI with Google’s Agent Development Kit (ADK) to orchestrate seven collaborating agents
  • AI Models: Vertex AI’s Gemini 1.5 Flash powers reasoning, summarization, and natural language generation
  • Data: CMS hospital cost data and BigQuery for large-scale querying
  • Streaming: Server-Sent Events (SSE) for real-time agent updates
  • Storage: Cloud Storage for generated reports and caching results

Each agent runs independently in a containerized ADK workflow, communicating through shared session state and event streaming.

[ \text{User Query} \rightarrow \text{ADK Agents} \rightarrow \text{BigQuery + Gemini} \rightarrow \text{Insight Report} ]

🚧 Challenges we ran into

  • Multi-agent orchestration: Synchronizing seven concurrent AI agents required careful event handling and state management.
  • Data normalization: CMS datasets are massive and inconsistent; mapping DRG/CPT codes to plain-language queries was complex.
  • Latency optimization: Achieving under 5 seconds total response time demanded asynchronous data fetching and parallelized agents.
  • Frontend streaming: Building a smooth real-time UI with SSE and animations was tricky, especially when aligning with the backend’s sequential events.
  • Insurance cost estimation: Translating average payer data into realistic patient-level out-of-pocket predictions required careful reasoning models.

🏆 Accomplishments that we're proud of

  • Built a fully functional AI care team that collaborates live, each with specialized roles.
  • Achieved end-to-end latency under 5 seconds, thanks to serverless scaling and event-driven design.
  • Created a visually engaging, educational frontend where users see AI agents explain complex healthcare costs.
  • Integrated Gemini 1.5 Flash for human-readable cost summaries.
  • Deployed both frontend and backend on Cloud Run with secure, scalable HTTPS endpoints.

🧠 What we learned

  • How to build and orchestrate multi-agent systems using Google’s Agent Development Kit (ADK).
  • How to combine LLM reasoning (Gemini) with structured cost data (BigQuery) for explainable AI.
  • How critical UI design is in making AI feel transparent and trustworthy.
  • The importance of clear state management in streaming agent workflows.
  • The potential of AI in healthcare equity — making cost data understandable for everyone.

🔮 What's next for CareCostAI

  • 🌍 Geo-aware hospital discovery: Automatically find and compare hospitals near the user.
  • 🧾 Insurance integration: Pull real coverage data from payer APIs.
  • 📊 Trend dashboards: Build cost insights using Looker or Data Studio.
  • 🧬 Patient profile personalization: Tailor cost predictions based on health, demographics, and insurance.
  • 🏥 Expansion to outpatient, dental, and pharmacy data.
  • 🤝 Partnerships with insurers to pilot real-world transparency tools.

Our vision is to make healthcare costs as easy to understand as a restaurant menu — empowering patients with clarity, confidence, and choice.

Built With

Share this project:

Updates