I am building Aurora, the world’s only truly Autonomous Algorithmic Trading Agent
Building an Orchestrating AI Agent for Advanced Financial Research and Analysis
When I was a software engineer at Oscar Health, I built an AI Agent before it was cool.
I was on the Oscar Claims Engine and Logic team (Ocelot). Ocelot built out Oscar’s internal claims adjudication system, which is responsible for generating over 95% of Oscar’s revenue.
After being inspired by a Hackathon, we built and productionized Ocebot, an orchestrating AI agent that could understand the decisions produced from our claims processing system far better than a human processor.
This experience taught me the difficulties in building AI agents. It also taught me its potential. Having successfully built an AI Assistant for algorithmic trading, I thought to myself… what if I could use this assistant more autonomously?
And thus, the idea for Aurora 2.0 was born. While Aurora can already help traders streamline financial analysis and deploy algorithmic trading strategies, I’m empowering her with an agentic brain to make her fully autonomous.
Here’s how she will work.
Context: What have I already built?
Press enter or click to view image in full size The UI for Aurora, the AI Assistant in NexusTradeTo fully understand the potential of Aurora, allow me to recap what I’ve already built.
For the past four years, I’ve been building NexusTrade, an AI-Powered algorithmic trading platform.
NexusTrade empowers traders to make smarter, more profitable trading decisions. At the heart of NexusTrade is Aurora, an AI assistant capable of screening for stocks, analyzing earnings, and backtesting algorithmic trading strategies, all without you having to worry about writing a single line of code.
I’ve used Aurora to great success, gaining more than $30,000 in the past year.
Press enter or click to view image in full size A screenshot of my Robinhood account balance – Up $29,806 to a current account balance of $50,850. I described how I used Aurora to gain this much in this articleBut using Aurora is a challenge for most investors. Most ordinary people don’t understand the process of performing research and transforming that research into strategies. When an error occurs, they don’t know how to fix it.
More importantly, even with an AI assistant, traders still need to:
- Know what questions to ask
- Understand how to interpret results
- Connect multiple analyses together
- Iterate on strategies manually
An AI agent avoids this entirely by handling the entire research process autonomously.
Technical Deep Dive: Turning an Assistant into an AI Agent
An AI agent is able to autonomously answer extremely complex financial questions. It’s an upgrade from an assistant because it is able to act autonomously and perform advanced analysis before answering your question.
Autonomously creating trading strategies
For example, let’s take an example of creating an algorithmic trading strategy:
Create a strategy that rebalances between UPRO and GLD. Figure out the best allocation split and rebalancing frequency
Press enter or click to view image in full size Asking the current generation of Aurora to create a trading strategyIf you ask Aurora right now, you’ll get an answer like this.
Of course. I can help you set up a rebalancing strategy between UPRO (ProShares UltraPro S&P500) and GLD (SPDR Gold Shares).
However, I cannot determine the “best” allocation split or rebalancing frequency for you. The optimal strategy depends heavily on your individual risk tolerance, investment goals, and market outlook.
Could you please tell me what allocation split you’d like to use (for example, 60% UPRO and 40% GLD) and how often you’d like to rebalance (for example, monthly, quarterly, or based on a specific market trigger)? Once you provide those details, I can create the portfolio for you.
The reason for this is simple – right now, Aurora acts with a single request/response relationship. While she has the right tools to create and test different trading strategies, it’s not possible to actually accomplish the user’s goal with a single request. We need an iterative process of creating strategies, testing them, and evaluating them, all autonomously.
Here’s how we can accomplish this.
Creating an orchestration prompt
Press enter or click to view image in full size The Orchestration Prompt for AuroraTo create a fully autonomous AI agent, I need the AI to have the ability to reason about the stock market and make actions based on what it finds. Luckily, there’s a framework that allows you to do this called ReAct.
No, I’m not talking about the frontend programming language. ReAct stands for ‘Reasoning and Acting’. It’s a framework that enables AI agents to iteratively reason through complex problems and take actions based on their observations.
Essentially, it’s the secret that transforms a simple chatbot into an autonomous problem-solver capable of handling multi-step analysis. Using the orchestration prompt and the same question above, a response from the ReAct would be the following.
{
"thought": "The user wants to find the optimal rebalancing strategy for a
portfolio of UPRO and GLD. This involves testing different allocation
splits and rebalancing frequencies. I will need to create multiple
portfolios, backtest them, and then analyze the results to determine
the best combination.
My plan is as follows:
1. **Create Portfolios**: I will start by creating a set of portfolios
with varying allocation splits (30/70, 40/60, 50/50, 60/40, 70/30) and a
fixed rebalancing frequency (e.g., 30 days).
2. **Backtest Portfolios**: I will then backtest these portfolios over a
significant historical period.
3. **Repeat for Different Frequencies**:
I will repeat steps 1 and 2 for other rebalancing frequencies (e.g., 60
days, 90 days).
4. **Analyze Results**: Finally, I will analyze all the backtest results
to identify the best-performing strategy based on metrics like Sharpe
ratio, total return, and max drawdown.
I will begin by creating the first batch of portfolios with a
30-day rebalancing frequency.",
"action": {
"tool": "Create Portfolios V2",
"input": "Create 5 portfolios that rebalance every 30 days
between UPRO and GLD with the following allocation splits (UPRO/GLD):
30/70, 40/60, 50/50, 60/40, and 70/30."
},
"requiresMoreSteps": true,
"finalAnswer": ""
}Then, we take the input and the tool and send it to the AI like a regular message. After the response, we input that as the next message in the ReAct framework, and we iteratively chat with the AI until `requiresMoreSteps` is false.
Press enter or click to view image in full size The next step in the iterative ReAct processThis can be a long process, but the end result is well worth it. We will have an accurate answer to our to tough financial question.
Press enter or click to view image in full size The final answer from Aurora. She By running this iterative process for “Create a strategy that rebalances between UPRO and GLD. Figure out the best allocation split and rebalancing frequency”, we are able to make several conclusions, including that more frequent rebalancing tends to lead to better returns, and that the best UPRO/GLD split tested was the 50/50 split.
This is just one example of how an autonomous AI agent can streamline financial analysis. Using the same framework, we can answer questions such as:
- What type of sentiment is associated with a small cap bio stock before it moves 50% in a month?
- If I’m creating a rebalancing strategy for large cap stocks, what are some filters I should include?
- What is the best SMA to look at when looking to “buy the dip”?
Using NexusTrade, you can already find the answers to these questions by sitting down and doing research. But after agentifying Aurora, you won’t need to. She’ll be able to answer the toughest financial questions fully autonomously.
How insane is that?
What’s so unique about this?
The reason why this is so powerful is that this technology doesn’t exist in the retail space.
While there are are many AI-powered trading tools available, most fall into these categories:
- No-Code Trading Bots: Execute rules automatically, but are simplistic and require a separate platform for research
- AI Assistants: Answer questions and provide analysis (like ChatGPT and Claude)
- Robo-advisors: Automate portfolio management with fixed strategies (like Betterment or Wealthfront)
- Charting software: Helps with charting and analysis, but not real-world trading and research
What’s missing from ALL of these? The ability to autonomously:
- Take a complex question like “find the best rebalancing strategy”
- Break it down into research steps
- Execute those steps iteratively
- Learn from results and adjust approach
- Deliver a complete, tested solution
Aurora is the first AI that is even able to do all of this. It is the world’s only publicly-available truly autonomous algorithmic trading agent. This is not a bold, unsubstantiated claim – I’ve done my homework and I challenge you to prove me wrong.
Press enter or click to view image in full size A table showing competing solutions to NexusTrade – NexusTrade’s Aurora is the only one that can autonomously research, test multiple combinations, and deliver an optimized solution without human interventionThis is the first time that retail will have Wall-Street grade tools and the ability to use them at their fingertips. This can change investing as we know it.
I’m not exaggerating.
Concluding Thoughts
When I was a software engineer at Oscar Health, the AI agent I helped build saved Oscarians thousands of hours and reduced the need for highly technical operations employees to understand claims processing outcome.
This same impact can be applied to algorithmic trading.
Even with AI tools, algorithmic trading remains inaccessible to all but the very advanced. But not for long. I’m building Aurora to be the world’s only truly autonomous AI agent for financial research and analysis. Investors of all experience levels will be to create algorithmic trading strategies starting from an idea.
And that’s just the beginning.
Imagine a world where AI creates dozens of portfolios, continuously tests them, and even manages them for the long run. That world is suddenly within reach. Will you be like 99% of others and be left behind for the AI revolution? Or will you join me and the other data-driven investors and embrace the true power of AI?
The choice is up to you.
