AutoClaim

Inspiration

Our team was inspired by ongoing challenges in U.S. healthcare payer systems—manual reviews, delayed claim processing, high administrative costs, and lack of transparency.

We drew particular motivation from one of Gartner's tech trend, which emphasize that U.S. healthcare payer CIOs are investing heavily in AI-first data management strategies. This validated our vision of building an agentic AI system that blends deterministic accuracy with the adaptability of LLMs to modernize healthcare claim adjudication.

We wanted to prove that agentic AI can deliver real business value by reducing inefficiencies, lowering costs, and improving both patient and provider experiences.

What it does

HealthcareAgent is a hybrid claims adjudication platform that combines: Rules-based processing for deterministic, policy-aligned decisions AI-driven adjudication for ambiguous or complex claims Transparent explanations that cite relevant policies, provide reasoning, and offer confidence scores Human-in-the-loop workflows for final oversight where necessary Analytics dashboards for tracking efficiency, trends, and compliance At its core, the system learns continuously from processed claims, improving accuracy and speed over time.

How we built it

We designed AutoClaim with a modular architecture to balance scalability, transparency, and AI integration: Architecture Overview (Raw Claim→Rules Engine→Ambiguity Detection→RAG Context Retrieval→LLM Adjudication→Human Review (if needed) Raw Claim→Rules Engine→Ambiguity Detection→RAG Context Retrieval→)zLLM Adjudication→Human Review (if needed))

Backend API Layer (FastAPI): Provides REST endpoints with auto-generated OpenAPI documentation Handles rules processing, data pipelines, and ML/LLM integrations Frontend (React + Mantine UI): Displays claims, adjudication results, and review queues Offers interactive dashboards for analytics and compliance monitoring Databases: SQLite for structured claim data and state ChromaDB (Vector DB) for semantic search, precedent retrieval, and policy grounding AI Pipeline: Hybrid Rules Engine (deterministic logic for simple cases) RAG Layer (retrieves similar cases and policies) LLM Adjudicator (Gemini API) that generates policy-grounded recommendations

Challenges we ran into

  Balancing automation with oversight: Ensuring the AI didn’t overstep in ambiguous cases while still saving human reviewers time.
 Hallucination risk: Avoiding ungrounded LLM outputs required a carefully designed RAG pipeline.
 UX design: Building an interface that presents dense claim and policy information clearly to reviewers.

Accomplishments that we're proud of

 Built a working hybrid rules + AI adjudication pipeline that processes claims end-to-end.
 Implemented a priority-based human review queue, improving efficiency.
 Created an analytics dashboard that visualizes cost savings, throughput, and compliance trends.
 Demonstrated a realistic ROI model, showing potential $1.5M annual savings for a mid-sized payer.

What we learned

Healthcare The intricate relationships between CPT procedure codes and ICD-10 diagnosis codes The complexity of medical necessity determinations How modifier codes substantially alter the interpretation of procedures The nuanced differences between different healthcare plans and their coverage policies Technical Skills Implementing a hybrid rules engine that blends deterministic rules with probabilistic AI judgments Building RAG (Retrieval-Augmented Generation) systems that ground LLM outputs in factual data Creating effective prompts that instruct LLMs to perform specialized tasks while avoiding hallucinations Managing the complex state and UX requirements of a claims review interface System Design Balancing automatic processing with human oversight Creating transparent AI systems that provide explanations for their decisions Designing interfaces that present complex medical and billing information in an accessible format Implementing efficient database schemas for specialized healthcare data

What's next for AutoClaim

Integration with EHR systems: Connect with existing healthcare IT infrastructure via HL7/FHIR APIs Multi-payer adaptation: Expand rules engine to support varied coverage policies across insurers Advanced analytics: Predict appeal likelihood and proactively flag at-risk claims Regulatory certification: Ensure compliance with HIPAA and CMS guidelines Scalability: Deploy at enterprise scale with containerized infrastructure (Docker + Kubernetes)