# Aegis
Underwriting teams receive messy, multi-format claim packs. The goal is to convert those artifacts into an auditable, typed case summary with linked evidence so reviewers can reason faster and more consistently.
Overview
Aegis is built as the document-intelligence companion to ParaEval. Its job is to turn claims material into typed evidence, retrieve corroborating context, and produce a recommendation that stays transparent about uncertainty. OCR and visual reasoning convert uploaded PDFs and screenshots into typed facts. Each case maps extracted facts to policy thresholds, permitted evidence types, and review notes so the recommendation stays grounded in contract context. Aegis retrieves sensor logs, field reports, and modeled outputs that help an underwriter decide whether the uploaded document is sufficient, corroborated, or weak. The immediate output is a recommendation, not a hidden final answer — when the case becomes trigger-sensitive, the file can hand off to ParaEval for stricter adjudication.
Problem
Scanned PDFs, photos, and dashboard screenshots resist search and comparison, and rarely map cleanly to policy clauses. This creates manual re-keying, fragmented provenance, and variability in how similar cases are assessed.
Approach
A modular intake pipeline that chains OCR (DeepSeek), visual reasoning (Qwen3-VL), local retrieval over uploaded artifacts, and a typed report generator. The Next.js app exposes both mock and live paths so the end-to-end flow is reproducible in demos and safe without external credentials.
Impact
- * Produces a structured, citation-rich memo with links back to the original artifacts for auditability.
- * Reduces copy-paste by normalizing extracted facts into a typed schema that downstream tools can consume.
- * Pairs naturally with ParaEval: Aegis extracts and justifies evidence; ParaEval adjudicates formal trigger decisions.
Architecture
Aegis is built as the document-intelligence companion to ParaEval. Its job is to turn claims material into typed evidence, retrieve corroborating context, and produce a recommendation that stays transparent about uncertainty.
Guardrails
- * Keep PII scrubbing explicit before wider retrieval or escalation.
- * Separate document understanding from trigger adjudication so each step remains explainable.
- * Expose evidence quality and missing data rather than pretending every upload is decisive.
- * Gate live parsing and model calls behind environment configuration in production.