I am a lawyer, and I use Claude Code for 10 hours a day to automate accounting firm workflows and support law office operations. I am building a RAG pipeline to be able to process 40,000-page investigation file into a vector database and support defense work. The system is self-learning and schema-evolving. I am convinced that AI-based process-support systems should be introduced across the entire public administration. In legislation, whenever I run my systems on a law, I find critical errors within minutes—so the entire legislative and legal drafting process needs to be transformed. Based on my experience so far, I believe I am capable of building a RAG system that can detect abuses from the past 16 years through AI processing of invoices, contracts, and performance reports. I can only imagine what an expert team could achieve. Claude for Government exists, and I think mistral.ai could process 1,000 pages for the state for about 10 cents if contracted. Right now, I am building a Karpathy-style knowledge base. The pipeline was completed in one hour, and it will organize my knowledge and new information into a wiki-like database. The system feeds me filtered new knowledge on a daily basis, relative to my existing wiki-style knowledge, and does not burden me with already known information. Since last November, the possibilities seem limitless to me.