“I am not a software engineer. I am a trader who ‘vibes’ with AI.”
When I decided to move from manual scalping to automated swing trading, I faced a massive wall: Programming.
I needed a dashboard to monitor real-time prices, calculate complex indicators like Volumatic VIDYA, and execute trades automatically 24/7. But my Python skills were non-existent.
Instead of spending 6 months learning syntax, I built a professional-grade system in just a few days using what I call the “Google Trinity”: Gemini 1.5 Pro, NotebookLM, and Google Antigravity.
This post is not about “how to code Python.” It is about how to conduct an orchestra of AI tools to build something beyond your own skill level. Here is the deep dive into my workflow.
1. The Roles: My AI Development Team
I treat these three tools not as software, but as my employees. Each has a specific job description.
🧠 Gemini 1.5 Pro: The Senior Lead Developer
Gemini is the engine. But unlike other LLMs, Gemini 1.5 Pro has a massive Context Window (up to 2 million tokens).
- Why it matters: Other AIs “forget” your previous code as the conversation gets long. Gemini remembers my entire project structure—all 10+ Python files, CSS styles, and strategy logic—at once.
- My Command: “Here is my current main.py and strategy.py. I want to add a ‘Stop Loss’ feature that triggers at -2%. Rewrite the code to integrate this safely.”
📚 NotebookLM: The R&D Department (No More Hallucinations)
Coding AI often lies (hallucinates) about library functions. NotebookLM solves this by “Grounding.”
- The Problem: Gemini might invent a fake function for the exchange API.
- The Solution: I upload the official PDF documentation of the exchange API (e.g., Upbit, Binance, CCXT) into NotebookLM.
- The Workflow: When I need to know how to fetch “Order Book” data, I ask NotebookLM first. It gives me the exact, fact-checked code snippet based on the docs. I then feed this correct info to Gemini.
🚀 Google Antigravity: The Agent “Mission Control”
This is the game-changer. It is not just a text editor; it is an “Agent-First” IDE where I command AI agents to work for me.
- The Role: Unlike traditional VS Code where I type code, in Antigravity, I assign tasks to autonomous agents.
- The “Magic” Feature: The Browser Agent. I can tell Antigravity, “Open the dashboard in the browser and test if the buy button works.” The agent actually opens a Chrome window, clicks the button, and reports back if it failed. It brings the code to life instantly.
2. The “Vibe Coding” Workflow: A Real Example
How do these three actually work together? Let’s look at how I built the “RSI Divergence Detector” feature.
Step 1: The Idea (Vibe)
I didn’t write “def calculate_rsi():”.
I simply wrote to Gemini:
“I want to find stocks where the price is falling, but the RSI indicator is rising (Bullish Divergence). Create a function to detect this.”
Step 2: The Fact-Check (NotebookLM)
Gemini needs to know how to get the data first. I went to NotebookLM (where I uploaded the CCXT library manual) and asked:
“How do I fetch the last 100 candles for BTC/USDT using CCXT?”
NotebookLM Answer: exchange.fetch_ohlcv(‘BTC/USDT’, timeframe=’1h’, limit=100)
Step 3: The Synthesis (Gemini)
I went back to Gemini with the verified code from Step 2:
“Okay, use this fetch_ohlcv code to get data, and then apply the RSI Divergence logic you mentioned earlier. Visualize it on a Streamlit chart.”
Step 4: The Execution (Google Antigravity)
Gemini generated the full script. I pasted it into Antigravity.
- Error: IndexError: list index out of range.
- Fix: I pasted the error message back to Gemini. “Fix this.”
- Result: 10 seconds later, the dashboard was live in the preview window, showing the chart perfectly.
3. Why Not ChatGPT or Claude?
I have used them all. But for System Building (not just writing one snippet), Google wins for two reasons:
- The “Big Picture” Memory: As mentioned, Gemini’s context window is a game-changer. I can upload my entire backup file. It understands how a change in file A affects file B. ChatGPT often loses track of the file structure in long chats.
- The Ecosystem: Moving data from Drive (API Docs) to NotebookLM, and copying logic to Gemini, feels seamless.
4. Advice for Aspiring Vibe Coders
You don’t need to study syntax. You need to study Logic and Communication.
- Don’t ask: “How do I write a for-loop?”
- Ask: “Compare the volume of the last 5 minutes with the average of the last hour. If it’s 500% higher, send me a Telegram alert.”
The Google Trinity handles the “How.” You just need to focus on the “What.”
This dashboard is just the beginning. In the next post, I will share the specific “Prompts” I use to debug errors without looking at a single line of code.