Patent Draft Generator (LLM)

A local-first patent drafting system built on Ollama (Mistral, Mixtral 8x7B, Llama 3.1 70B). It takes an invention disclosure or rough notes and generates a draft specification: title, background, summary, claims, and abstract.

• Config-driven prompt architecture with patent-office rule injection for USPTO, EPO, and IPO
• Stop-sequence guards and validation checklists to keep output in the required structure
• Structured evaluation harness to compare models and prompt versions
• Containerized and deployed on Google Cloud Run