Patent drafting demands the identification of technical limitations in prior art and the formulation of novel, patentable solutions, a process traditionally reliant on extensive domain expertise and manual effort. To overcome this challenge, we propose iPatent, a novel multi-agent framework that integrates Large Language Models (LLMs) with Theory of Inventive Problem Solving (TRIZ) to automate patent ideation and documentation. iPatent orchestrates four critical stages: (1) prior art analysis using LLM-driven semantic search to extract technical pain points and map them to TRIZ contradictions; (2) innovation generation via TRIZ-based contradiction resolution enhanced by dynamic domain adaptation; (3) compliance-aware patent drafting that synthesizes the analysis and novel solutions into structured disclosures; and (4) hybrid quality assessment combining rule-based checks with LLM-powered scoring. Evaluations across diverse technical domains demonstrate that iPatent delivers precise identification of technical contradictions and generates solutions recognized by experts for their novelty and practical viability. The system significantly streamlines the patent drafting workflow while rigorously adhering to statutory documentation requirements, bridging AI-driven automation with systematic innovation methodologies to offer a scalable solution for intellectual property management.