solo-developer

Startup using agents for SaaS feature request triage

CrewAIDocling OCRLlamaIndex RAGOpenAI GPT-4oStreamlit
Stack tools5
AddedMar 2026
StatusPublished

Demo processed property damage claim approving $8,500 roof repair, rejecting fence/tree damage, generating investigation tasks and justifications; enables handling 100x more claims

solo-developer

Why they built it

To augment insurance teams by automating claim evaluation against policies, saving hundreds of hours yearly and enabling faster handling of more claims.

What worked

Docling preserved PDF visual hierarchy for accurate section queries; RAG provided auditable policy excerpts; agents collaborated sequentially to produce structured outputs.

What broke or was painful

Unstructured PDFs lost layout/info (used Docling workaround); LLM hallucinations on damage semantics (prompt refinements)

The result

Demo processed property damage claim approving $8,500 roof repair, rejecting fence/tree damage, generating investigation tasks and justifications; enables handling 100x more claims

References