insurance_nerds

InsuranceNerds is a chat interface built for Cornerstone Insurance Agency's customer service employees. It uses retrieval-augmented generation (RAG) to answer questions about agency procedures, policies, and documentation stored in Microsoft 365 SharePoint. The system syncs documents from SharePoint, chunks and embeds them into pgvector, and uses OpenAI GPT-4o to generate grounded responses. Admins can manage canonical answers, monitor conversations, review feedback, and control which documents are synced.

Built to solve a real business problem at Cornerstone Insurance Agency: CSRs need fast, accurate answers to procedural questions without hunting through SharePoint folders or interrupting colleagues. The system grounds AI responses in actual company documentation to reduce hallucinations and ensure answers reflect current policies. The extensive brand guidelines, voice instructions, and admin tooling suggest this was developed with a specific client deployment in mind.

Laravel 11 (PHP 8.3), Livewire 3 for reactive UI, Blade templates with Tailwind CSS, MySQL for app data, PostgreSQL 16 with pgvector for semantic search, Redis for cache/queue, OpenAI GPT-4o and text-embedding-3-small, Microsoft Graph API for SharePoint sync and OAuth authentication, Docker with Dokploy for deployment, PHPUnit for testing, CI via GitHub Actions.

Last active Apr 8, 2025 — 126 commits across 15 active days (spanning 99 days).

Development spanned 99 days (late December 2025 to early April 2026) with 126 commits across 15 active days. The bulk of the work happened in three intense bursts: initial build (38 commits in December), core feature development (41 commits in January), and a major push in March (39 commits). This uneven pattern suggests focused sprints rather than daily iteration—likely driven by client feedback cycles or deployment milestones.

This is clearly a shipped production application for a real insurance agency. Evidence: detailed brand guidelines based on cornins.com colors, highly specific voice instructions for CSR interactions, comprehensive admin documentation ("For Laura and Admin Users"), production deployment workflows (staging + prod GitHub Actions, Dokploy configs), Microsoft OAuth integration with tenant-specific setup, and granular features like conversation monitoring, feedback loops, and canonical answer management. The polish and specificity far exceed a prototype.