Hi Guys, Emma here from the U.K. starting out developing code for some automated trading to back test a particular strategy. has anyone any experience of using Claude sonic/sonnet for generating codes from a detailed description of trade set up and through to execution. Is this really a thing thats possible.
Hi Emma, welcome to the community!
Great question and yes, using Claude, ChatGPT or similar AI tools to generate ProBuilder code is absolutely something people are doing, and it can give you a decent starting point, especially for straightforward logic like basic indicator crossovers or simple entry and exit conditions.
That said, it is never fully reliable on its own, and here is why. General purpose AI models have a broad but surface level knowledge of ProRealTime and ProBuilder. They tend to struggle with the more subtle aspects of the platform, things like how specific order types behave, broker side constraints, the way certain functions interact with live data versus historical data, or the precise syntax required for some of the less common instructions. You can end up with code that looks right but throws errors, or worse, code that runs but does not do what you actually intended. In a backtesting context that can give you misleading results, which is the last thing you want when you are trying to validate a real strategy.
So the best approach right now is to use AI as a rough drafting tool if you like, but always bring the result here to the forum to get it reviewed by people who actually know the platform inside out 😉
Post your trade setup description, share what you are trying to achieve, and the community will help you get to something solid and tested.
On a related note, we are currently building our own dedicated AI assistant specifically trained on ProRealTime programming, and it will be a completely different experience compared to generic models. The reason for that is simple: ProRealCode is the reference source that most of those general AI models have drawn their ProRealTime knowledge from in the first place. Our own tool will go much deeper and stay accurate on the specifics that matter. We will share more on that soon.
In the meantime, feel free to post your strategy details and we will get you pointed in the right direction.
OK this is a fabulous response, including my suspicions being addressed, so thank you very much. Ive been a bit lazy when it comes to backtesting as its a full process in itself and although i don’t want to cut corners , its will be wise to have useable indicators to at least trigger sets ups in place. yes i also see how the syntax is dealing 9 or rather not dealing) with the Claude generated codes, which is why i have stripped it back to creating an indicator only rather than a fully strategy set up…AI simply couldn’t cope with the executions…was firing up far too many errors. So yes thank you for this and i shall for certain flick over the indicator once its working for it to be tested by the community. Regards.
Hi Emma!
As Nicolas rightly says, business models have a superficial knowledge of Probuilder and you have to be careful because they mix languages… That said, here’s my opinion on business models.
I’ve thoroughly tested ChatGPT, Gemini, and Claude, using the top versions of each. The best by far is Claude, specifically the Opus 4.6 model.
I hope this helps. We’re here to help 🙂
thank you, ive started working with claude. im feeling im about to level up , after 7000 hours already studying and participating! thanks for the support guys.
Thank you for opening this discussion on the use of AI with ProRealTime.
AI is a fantastic tool for answering our questions, and I personally use it from time to time for coding. Regarding previous discussions, I would like to contribute the following points:
· Dedicated Section: Perhaps in the future, we should have a discussion category specifically dedicated to AI in ProRealCode. This would prevent interesting conversations from getting lost among other topics.
· Code Mastery: I believe we must always remain masters of our own code. AI should not be used to code in place of the user. Mastering the code allows the user to modify and structure it so that it can be easily and properly integrated into other indicators and/or screeners.
· Value Add: Using AI to manage code based on standard indicators (MACD, RSI, etc.) is of limited interest, as the ProRealCode library already contains many examples that meet these needs. However, AI is excellent for generating a coding draft for new, original algorithmic ideas that might exceed one’s own implementation knowledge—such as using specific mathematical formulas or principles.
· Prompt Precision: When making a request to an AI, I recommend describing the requirements as precisely as possible from the very start. It is best to avoid multiple iterations with new criteria or changes to the initial scope, as this risks generating complex and potentially inappropriate code.”
“Translated from French with Gemini”
