Iris

What We Built

Iris is an AI-powered pre-submission claims validator. India's cashless healthcare system has a major problem—patients discover errors in documentation after rejection, when it's too late - pressuring both patient and the hospital. Iris steps in to validate the documentation before insurers see your claim. Pre-Authorization Validator employs a multi-agent architecture: completeness checker validates fields against policy schemas, policy compliance engine cross-references waiting periods and coverage limits, medical necessity analyzer evaluates treatment-diagnosis alignment, and fraud/waste/abuse detector performs context-aware anomaly detection. Outputs: readiness score (0-100) with granular recommendations and plain-language patient summaries. Discharge Validator performs intelligent bill reconciliation through cost variance categorization, employs an LLM-based escalation analyzer determining if variances are medically documented, and generates professional recovery instruction PDFs with schedules, follow-up timelines, and activity guidelines. Key Innovation: Range-aware validation performing contextual risk assessment against procedure-specific cost distributions, hospital tier adjustments, and complexity factors. Multi-modal reasoning distinguishes legitimate clinical variations from documentation gaps.

How We Used Claude

Claude Sonnet 4.5 powers the entire validation pipeline. We leverage Claude's nuanced reasoning for distinguishing legitimate complications from fraud patterns, and multimodal capabilities for intelligent PDF extraction when structured parsing fails. Beyond its implementation in the tool, Claude's coding abilities were instrumental in building Iris itself—generating complex Pydantic schemas, architecting the multi-agent orchestration layer, and implementing sophisticated prompt engineering patterns.

Impact & Future

Hospitals reduce rejections, patients eliminate surprises, discharge times drop from hours to minutes. The future is learning and proactive: analyzing historical outcomes, suggesting improvements pre-submission, creating feedback loops from insurer queries. Recovery instructions could evolve into a proactive follow-up companion—activity reminders, automated scheduling etc. For insurers: standardized submissions enable faster adjudication and sophisticated fraud detection. Vision: Iris as infrastructure—the validation layer India's cashless system needs, making transparent, anxiety-free healthcare standard practice.

Built With

• claude
• pydantic
• python
• regex
• streamlit