An ex-Jane Street trader told me the secret to AI trading. So I built it.
Episode 2 of the $25,000 Public Portfolio Challenge
If you ever watched “The Big Short”, you know the type. “My quantitative.” A smart, retired millionaire in his early 30s who’s scary good at math. This ex-Jane Street guy fit the bill to a tee.
Let’s call him “X”.
X and I were talking about applications of AI in trading. We were sharing our experiences, what we were building, and our performance. His prototype shocked me – up more than 20% in the past 3 months.
I asked him his secret.
“What would you do if you were in my shoes?”
He told me.
So then I built it.
A Callback to my $25,000 Public Portfolio Challenge
This challenge is something I encourage anybody with guts to do.
Deposit $25,000 into a Public portfolio.
Public here actually means two things. Number 1, it refers to the brokerage. It’s like Robinhood but built for community building and serious traders looking to find their edge. For deploying $25,000, they are my trusted partner.
But Public also means public, like the opposite of private. Because you can see my real-time trading performance at all times.
Press enter or click to view image in full size My Public Account: Starting with an initial value of $25,000. Check it out yourself hereTo start the challenge strong I did a weekend sprint to implement the core features to transform my agent into an autonomous system that can run 24/7 (or whenever I want). This directly leverages the advice I received from X last week.
Here’s what he told me, and how I applied it.
The Secret
“It’s shockingly easy now,” X started. “When you claim to make such a product, the VCs want you to prove it. So with Cursor and Claude I worked on the algos little-by-little.”
He went on, “the initial strategies from the LLMs were like really bad. You know, what you would find online. So I prompted it to do better. Do x, not y. Use this metric. And started building these instructions for the model to follow. Over time, it got better.”
In my mind, I related this to my own experience. I might ask Aurora to build a trading strategy, and it’d be some cookie-cutter simple moving average strategy that clearly doesn’t work. But as I distilled knowledge from successful backtests, the results became noticeably better.
X continued.
“Back in the day, the things I wanted to analyze would take me weeks. Now I can do it in minutes”.
As the words entered one ear, my mind’s eye fixated on what he meant practically.
I asked him outright.
“What would you do if you were in my shoes?”
He answered immediately. He didn’t even think.
“Don’t spend money on expensive proprietary datasets. It is a waste”.
I started to get tense, remembering my $2,000/month data subscription had just charged my card. It feels like it’s needed. I still think it is. But maybe he’s got a point?
He continued.
“Create a loop that gathers data and runs at the same time each day. Analyze the data to find patterns, then act on it.”
He made it sound like a Leetcode easy.
I asked him explicitly,
“What type of data? Reddit data? News?”
He kept it vague. “Online, unstructured data. Lean on it.”
He could sense my disappointment, so he continued.
“Like with a purchased dataset, you don’t know how they got the data, and if it’ll change in the future. But by using unstructured data, you have access to the whole internet. Like YouTube or TikTok.”
I was intrigued. He kept going.
“Google search trends. Trending tweets. The TikTok FYP. The amount of unstructured data on social media and news sites is astronomical. You can see what works and what doesn’t and build your strategies around that.“
He repeated.
“Don’t waste your money on expensive alternative datasets.”
I let the words sit with me and thanked him for his time. I closed my computer, made some coffee, and sat down to think.
I’ve been relying so much on backtesting, and so far it works great. But what if I could make it reliable? Incorporate what people are actually saying, and finding trends that are actually robust.
Ten minutes passed, and I came back to the real world. I faced my laptop, opened it up, and got back to work.
Here’s what I’ve been cooking.
A 24/7 Stock Market Research Agent
I already built Aurora, a powerful AI trading agent.
Aurora was like Claude Code for trading. You could use her to perform complex financial analysis or backtest trading strategies… if you were the driver.
X’s advice inspired me to think bigger. Make her the OpenClaw of trading.
Not because OpenClaw is better. Far from it. But because it uniquely had the ability to schedule tasks.
I eliminated the gap and created something far better.
Creating an agent to run every day
After my fourth cup of coffee at 1am Sunday, I finally pushed my last commit.
And then I launched my first production test.
Press enter or click to view image in full size The initial prompt I gave Aurora, the AI trading agentMy other agent runs tend to create trading strategies that rebalance my portfolio at perfect weights. This one is noticeably different.
This agent will perform deep research on the news and see what stocks are trending. Then, it will search up their fundamentals and make a judgement. Will the news have a real impact on earnings? Or are these trending stocks fundamentally sound?
The AI will run automatically, every day at 9am.
The daily research strategy was attached to my Public Portfolio.If you go to my portfolio, you can see the agent strategy yourself.
I’m going to use this agent to see what other AI stocks are trending. I have some ideas about what stocks to buy. Maybe VRT stock or some other gems I found while using Aurora. If you want to be the first to know, here’s how you can get alerts to your inbox.
How YOU Can Follow Along in Real-Time
You don’t have to patiently refresh Medium every 2 minutes to get an update.
You can follow along in real-time and hold me accountable.
- Just go to my NexusTrade profile
- Create a free account
- Click the “Follow” button on the top of the screen
By following me, you can get notifications and watch every single portfolio that I deploy. Keep me honest, call me out. If I try to lie, the world will know.
Let’s make trading gurus honest again.
My 24/7 Trading Agents are Finally Launched
Before, to use Aurora, you had to sit down and type the strategies.
Now the strategies can make themselves.
X didn’t give me a magic formula. He gave me an idea.
With it, I built a framework – a way to automate the collection of unstructured data, run it consistently, and let the patterns reveal themselves.
Monday at 9 AM, Aurora runs her first autonomous research cycle on my live $25,000 portfolio. Not paper-trading. Not “in theory”.
My actual account.
Follow along and hold me accountable.
Episode 3 is coming. What should I build for Aurora next?
