To be honest, I prefer “commanding” over “doing” things myself. That is exactly why I love Vibe Coding.
- To Gemini: “Establish the strategy.”
- To NotebookLM: “Read this massive amount of data and summarize the key points.”
If you ask, they do it. That’s all there is to it. However, just like managing people, you need to ask well (write good prompts) for your AI partners to help you with all their heart.
AI is a tool for everyone, but depending on how you use it, you must make it a partner exclusively for you. This is the core experience I’ve gained while conversing with AI to build my trading system.
Here is my workflow for taming these two beasts.
1. Gemini: Hiring Your Personal Stock Expert (Using Gems)
Beyond simple Q&A, Gemini has a feature called ‘Gems’ that allows you to create a persona that acts like your dedicated employee.
1-1. How to “Hire” an Expert via Gems Type this into the Gemini chat window:
“I want to hire a stock trading expert within Gems and assign them a mission. Please tell me how to set this up.”
Gemini will kindly explain the method. But here is the important part: Do not just follow the default settings. You must add your specific requirements and ask it to “upgrade into a better partner.” This detail makes all the difference.
1-2. The New Gems (with OPAL) There is talk about integrating with Google OPAL recently. To be honest, I am not quite sure about this yet (the results aren’t coming out as I want, probably because I don’t know how to use it well…). I leave this part for you to study and explore.
1-3. Open Your Private Office Once the setup is complete, you will see a Gem icon in the left sidebar. This is the “Stock Expert” you just hired. From now on, do not use the general chat; click this icon to ask questions. You will receive much more professional and consistent answers. If the expert’s answer is unsatisfactory? Don’t fire them. Just tell Gemini to “Modify the expert’s role” to fit your needs better.
2. NotebookLM: Preparing “Food” for Gemini
To give Gemini good instructions, you need good questions and high-quality context. The tool that handles this research for me is NotebookLM.
2-1. Sourcing Create a new note, then add all the URLs containing information you like—whether they are YouTube videos or blog posts found through search.
2-2. The Golden Ratio: 10 to 15 Sources If there are too few sources, the answer is weak. If there are too many, the AI gets confused. In my experience, adding about 10 to 15 URLs is the sweet spot. NotebookLM will automatically analyze and summarize the data.
2-3. Ask Relentlessly Don’t just read the summary and stop. Keep asking questions in the chat. “What are the risks here?”, “What are the exact indicator settings?” Keep asking until your curiosity is fully satisfied.
2-4. Save as Text Copy the entire chat history and the summary, then save it as a text file. This will be the raw material for the “Work Order” you will send to Gemini.
3. Final Step: Ordering the Development Guide
Now, the preparation is complete. I hand over the material analyzed by NotebookLM (the text file) to Gemini (my Gems partner) and issue the final command.
I use the following prompt to request a development guide:
[Prompt Template]
I am planning to do Vibe Coding using Antigravity. Please review the attached document and write a development guide based on the following 4 criteria:
- Separate UI-related content from Logic-related content clearly.
- Propose methods for acquiring stock analysis information and specific analysis logic.
- Organize all details mentioned in the attached document without omission, and merge duplicate information.
- Write in Markdown format so I can copy and paste it immediately.
When I do this, Gemini spits out a perfect “Development Blueprint” based on the information I spoon-fed it. All I have to do is look at it and say, “Good, let’s proceed.”
Note: This post does not explain the basic usage of Gemini and NotebookLM (like how to sign up or where the buttons are). It is difficult to explain in text, so it is much faster to search for and watch YouTube videos.