ClearCare

• 

  ClearCare Introduction

• 

  Front Page

• 

  Result Shown

ClearCare: Secure Agentic Health-Wealth Orchestration Inspiration Healthcare in the U.S. is a "black box" of complex clinical jargon and fragmented financial data. For Medicare beneficiaries, this complexity leads to financial toxicity—where patients defer life-saving care simply because they cannot predict the cost.We were inspired to build ClearCare to bridge the gap between clinical symptoms and financial clarity. We didn't just want to build another chatbot; we wanted to create an autonomous agent that treats medical privacy as a first-class citizen while using professional-grade analytics to navigate the labyrinth of Medicare costs. What it does ClearCare is a multimodal AI agent that acts as a patient's financial and clinical advocate.Symptom Triage: Users describe symptoms via text or voice; the agent determines urgency and suggests the correct care path.Insurance Intelligence: It "sees" insurance cards and extracts plan details securely.Cost Prediction: It queries a governed semantic data layer to provide real out-of-pocket cost estimates.Voice Guidance: It speaks to users with a calm, clear voice, making complex insurance data accessible to seniors. How we built it ClearCare is built on a stateful LangGraph orchestrator, moving from static retrieval to dynamic, self-correcting action.The Secure Brain (Aira API): We utilized the Aira Framework for high-integrity, medical-grade reasoning. Aira handles symptom extraction and triage, ensuring that clinical interpretations are grounded in medical logic rather than general-purpose LLM patterns.The Semantic Action Layer (Lightdash API): To ensure 100% accuracy in cost estimation, we integrated the Lightdash API. Instead of the agent "guessing" costs, it queries a governed semantic layer. This allows the agent to pull precise metrics like $Total_Claims$ and $Deductible_Remaining$ directly from our governed data models.Empathetic Output (ElevenLabs): We integrated ElevenLabs to provide a "human-in-the-ear" experience, converting dense financial data into empathetic, spoken summaries for users who may struggle with small mobile screens. Challenges We ran intoNatural Language to BI Translation: Teaching an agent to talk to a Business Intelligence tool like Lightdash required building a robust Model Context Protocol (MCP). We had to ensure the agent understood that "Am I covered?" translates to a specific metric filter in the data model.The Hallucination Barrier: In medical costing, a "hallucinated" price is a liability. We solved this by implementing a Self-Critique Loop. If the Aira agent detects a discrepancy between the extracted plan and the Lightdash analytics, it autonomously triggers a secondary search to resolve the conflict.Syncing Multimodality: Managing the latency between a vision request (Aira), a data query (Lightdash), and a voice stream (ElevenLabs) was an engineering hurdle. We utilized FastAPI's asynchronous task handling to keep the experience fluid. Accomplishments that we're proud of True Autonomy: We successfully built an agent that doesn't just "think"—it acts by querying real-world databases and verifying its own logic.The "Confidence Meter": We developed a live UI component that displays the agent's internal self-critique score as it improves its own response in real-time.Zero-Knowledge Triage: Implementing a triage system that can identify "Red Flag" symptoms and prioritize them over routine cost-searches without human intervention. What we learned The core of our learning was that Autonomy requires Governance. We learned that for an AI to "act" in a high-stakes environment like healthcare, it must operate on a Semantic Layer. By using Lightdash, we gave our AI a "source of truth" that prevents the hallucinations common in traditional RAG systems.We also calculated the Care Efficiency Score ($\eta$) to help our agent make decisions: $$\eta = \frac{\alpha \cdot Q_p}{(C_{est} + 1) \cdot \ln(d + e)}$$ Where: $Q_p$ is the Provider Quality Metric. $C_{est}$ is the Estimated Cost (calculated as $(B \times M_s) - D_{rem}$). $\alpha$ is the Aira-assigned safety weight. What's next for ClearCare We aim to integrate FHIR APIs for real-time Electronic Health Record access and expand our Lightdash models to include pharmaceutical data, allowing ClearCare to become a total "Health-Wealth" management suite for every Medicare beneficiary in the country.

Built With

• aira
• cms
• css
• elevenlabs
• fastapi
• lightdash
• medicare
• next.js
• npi
• supabase
• tailwind
• tavily