Management consultant Yuan Heng Fan ordered an M4 Max MacBook Pro with 64GB of RAM—a machine that ran him more than 6 million won (roughly $4,500). It was far more horsepower than anyone needs just to run PowerPoint and Excel. So, in the span of a single day, he used it to build a fully local AI assistant he calls 'NG-Notebook.'
The project tackles a fundamental dilemma of using AI inside a company. Uploading confidential data—M&A documents, financial statements, strategic plans—to ChatGPT is a security non-starter. And Google's NotebookLM doesn't natively support Excel (.xlsx) or PowerPoint (.pptx) files either. So is there a way to tap into AI's intelligence while keeping sensitive corporate data safe?
An AI Ecosystem That Lives Entirely Inside a MacBook
Fan's system rests on four components. The brain is a large language model (LLM), and for that he chose Ollama—a management tool that lets you run open-source AI models like Meta's Llama 3 directly on Apple Silicon. No cloud service is involved; all the 'thinking' happens inside the MacBook.
The nervous system is LangChain. LangChain doesn't reason on its own, but it serves as the plumbing that connects the AI brain to the other components. Through document processing and search, it lets the AI read and understand files. Finally, a web interface built with Streamlit acts as the chat window between the user and the AI.
The striking part is that the entire system came together in a single day. We've entered an era in which an ordinary business professional—not an AI scientist—can build a custom AI of their own.
The DIY Era of Enterprise AI
This project is more than a tech demo. It signals a fundamental shift in how companies approach AI.
Security and customization become the crux. General-purpose AI tools are powerful, but the limits of a one-size-fits-all solution are clear. The future of enterprise AI isn't reaching for an off-the-shelf chatbot. It's deploying secure, private, highly customized AI agents that can work with proprietary data and internal systems without the risk.
The ability to run AI applications on a high-performance local device like the M4 Max is rewriting the rules of data security. The guarantee that confidential information never leaves the machine becomes a decisive advantage in finance, law, and medicine.
From AI User to AI Creator
The lesson of Fan's experience is unmistakable: we need to move from simply using AI to building it.
The familiar phenomenon of 'shadow IT' is now surfacing in AI as well—think of employees turning to ChatGPT or Claude for work without official approval. But the next step demands something more: the capacity to build the AI tools that fit your own needs.
Tools like Ollama, LangChain, and Streamlit already exist. The barrier isn't technical—it's the willingness to try. If a consultant pulled it off in a day, why couldn't other professionals?
Even the hardware bet on an M4 Max with 64GB of RAM no longer looks excessive. A machine of this caliber can comfortably handle enterprise-grade AI workloads. Factor in the recurring cost of cloud APIs, and it may even be the more economical choice.
In the end, it comes down to one question: Will you remain someone who uses AI, or become someone who builds it? As Fan's one-day project proves, that line is closer than you think.
