Zack shares a practical method for fine-tuning a language model (LLM) using his LinkedIn posts, leveraging the Claude Code tool. The process begins with gathering training data. Using the Claude browser extension, he scrapes his LinkedIn posts and formats them into JSONL, a structured format for machine learning. Each post is organized with roles to distinguish between the system, user prompts, and the assistant's responses. He suggests collecting at least 50 posts for reliable results, noting that he managed to scrape 98 posts after some persuasion. After gathering the data, Zack emphasizes the importance of validating the JSONL format. He instructs users to drop the file into a project folder and use Claude Code to ensure the format is correct and to fix any issues. The next step involves fine-tuning the model using OpenAI, specifically the gpt-4o-mini, which is suitable for style transfer and cost-effective. The training phase is relatively quick, taking around 15 to 30 minutes. Once the model is trained, users can prompt it with topics, and it generates posts in the user’s voice, saving them as Markdown files. Zack also highlights some advanced tips. Once the setup is complete, users can have Claude generate posts, provide feedback, and refine the system prompts further. For personalized touch, he suggests creating a Claude.MD with specific instructions, such as always starting with an outline or including action items. The article concludes with Zack’s experience using this model for his first post, which was primarily AI-generated but edited by him. Several comments indicate interest and excitement about automating LinkedIn posts or repurposing old content using this method.