NEXSUS AI

Agent-Based Provider Data Validation & Confidence Scoring Platform

EY Techathon 6.0 – Detailed Solution Submission

---

🚀 Overview

NEXSUS AI is an agent-based, workflow-driven platform built to solve a critical problem in the healthcare payer ecosystem: inaccurate, outdated, and fragmented provider directory data.

Insurance providers and third-party administrators (TPAs) depend on provider data for compliance, operations, and patient experience. However, manual and periodic verification methods fail to scale.
NEXSUS AI enables a single, continuously validated source of truth using confidence-based automation and human-in-the-loop governance.

Video Demo: https://youtu.be/3Qp-TgRXheA?si=PRad0F4i0JifIEIu

❗ Problem Statement

Provider directories across major healthcare payers suffer from systemic data quality issues:

• 40–80% of provider contact data is inaccurate
• Manual verification achieves only 60–70% accuracy
• Fully corrected data can take 117–280 days
• Patients often visit providers who no longer accept their plan, causing delays and unexpected costs

These challenges arise from manual, fragmented, and non-scalable verification processes, increasing compliance risk (e.g., No Surprises Act) and degrading patient trust.

---

👥 Target Users

Primary Users

• Payer Operations Teams
• Provider Relations Teams
• Compliance & Quality Assurance Teams

Secondary Users

• Healthcare Providers (for direct verification in low-confidence cases)

---

💡 Solution Overview & Value Proposition

NEXSUS AI replaces slow, manual audits with continuous, automated provider data validation:

• Establishes a single source of truth for provider directories
• Uses agentic workflows to mimic real-world human verification steps
• Automatically flags outdated or inconsistent records
• Explains why data was changed using Explainable AI
• Escalates only high-risk records to human review or provider outreach

This approach reduces operational overhead while improving accuracy, consistency, and regulatory readiness.

---

⚙️ Methodology & Approach

The platform follows a sequential, deterministic agent-based workflow, orchestrated using LangGraph.

Workflow

1. Data Ingestion

   • NPI and basic provider details submitted via user action or scheduler

2. Asynchronous Processing

   • Tasks routed through Celery + Redis

3. AI Agent 1 – Data Validation

   • Cross-verifies data against NPI registry and trusted web sources

4. AI Agent 2 – Data Enrichment

   • Fetches licenses, affiliations, and credentials

5. AI Agent 3 – Confidence Scoring

   • Uses XGBoost to compute confidence score
   • SHAP provides explainability

6. Decision Routing

   • ≥ 88–90% → Auto-approved and updated
   • 60–90% → Acceptable, monitored
   • < 60% → Escalated to automated calling agent or human review

7. Directory Update & Reporting

   • Canonical database updated
   • Audit logs and validation reports generated

---

📊 Impact Metrics

Data Accuracy

• High-confidence provider records
• Reduced duplicate and inconsistent entries

Operational Efficiency

• Time to validate 100 records
• Reduced manual verification workload

Update Velocity

• Faster provider onboarding
• Faster update propagation

---

🧱 Technology Stack

Core

• Languages: Python, SQL
• Backend: FastAPI
• Database: PostgreSQL (canonical provider database)
• Agent Orchestration: LangGraph
• Async Processing: Celery + Redis

Machine Learning & AI

• Model: XGBoost
• Explainability: SHAP

Data Sources

• Web scraping (trusted provider websites)
• External registries (NPI, state licensing sources)

Infrastructure

• Docker-based containerization
• Cloud-ready, horizontally scalable design

---

📌 Assumptions & Constraints

Assumptions

• Initial deployment limited to one state (e.g., Massachusetts)
• Provider data available via public registries or websites

Constraints

• Variable data quality across directories
• Rate limits and compliance constraints
• Regulatory requirements for auditability

---

🔐 Security, Scalability & Governance

• Confidence-based escalation ensures human-in-the-loop control
• Full audit logs for all data changes
• Secure access and encrypted storage
• Horizontally scalable via asynchronous workers
• Modular agents allow easy extension to:
  • New states
  • New data sources
  • Additional agents (e.g., calling, retraining)

---

🧩 Components Demonstrated / Planned

• XGBoost-based confidence scoring engine
• Explainable AI layer for decision transparency
• Automated calling agent for low-confidence records
• Geospatial confidence map for provider reliability

---

🚧 Project Status

Prototype / Proof of Concept

Developed for EY Techathon 6.0, with core logic implemented in notebooks and workflow validated through architecture and flow diagrams.

---

👨‍💻 Team

Team Name: NEXSUS AI

• Aparimeya Tiwari – Database & Confidence Scoring (XGBoost)
• Harshalee Malu – Provider Ingestion & Directory Management
• Sahil Adit – Data Validation, Enrichment & Orchestration

---

📄 License

License to be added.