Ciridae began with on-site discovery at the restoration firm — time spent understanding the specific negotiation logic the CEO applied when reviewing proposals before encoding it as structured AI-executable rules. This knowledge-capture step was the foundational work that made downstream automation possible. The pipeline was built to ingest two document types simultaneously: contractor estimates and insurance adjustment documents. Every line item is classified, and the system applies the encoded negotiation rules to flag discrepancies — items where the contractor's scope and the insurer's adjustment diverge in ways worth challenging. The output is an annotated comparison document ready for human review before negotiation, not instead of it. The full solution deployed in 48 hours using the Ciridae platform, integrating Claude, OpenAI, and Gemini models. The system immediately outperformed the prior manual process — identifying discrepancies with greater accuracy than human review 10% of the time — and recovered an average of $500 more per proposal from the first week of use.