AmpyFin_v3.0.1

- Who I am: A little background about me as the owner of the project. I've always been interested in trading and always wanted to work on creating my own trading project. I had background in ML, so I decided to do was utilize this in trading. You might be wondering why I decided to make this open source. There's potentially a lot to lose, but I would beg to differ. - Why Open Source: From the moral standpoint, when I was in uni and wanted to code my first automated trading bot, I remembered there was practically no publicly available trading bot. It was mostly trading gurus promoting their classes to get money or their channel to get revenue. This was something I promised myself many years ago if I do create a successful trading bot I will open source it so other people can potentially use my project to create better trained models or projects. Another thing is opportunity. I was able to learn a lot from critique. I had one open source trading project before - which is now defunct - but back then I was able to meet different people with different background ranging from quant developers at respectable prop trading firms to individuals who were just interested attending the same class as me. This interaction allowed me to learn what aspects I needed to improve this project on as well as learn new strategies that they used in their pilot / research programs. That's what's special about open source. You get to meet people you never thought you will meet before the project started. - What My Project Does Most prop trading firms / investment companies have their own ML models. I don't claim that mine is better than theirs. To be honest, we are outperforming a vast majority of them at the current moment (there are 6000+ trading firms we are tracking in terms of their portfolio). This is only 2 months since it's gone live so that might mean nothing in the grand scheme of things. Backtesting results for v3.0.1 showed favorable results with Max Draw-Down at 9.29%, R ratio at 1.91, Sortino at 2.73 and Sharpe ratio at 2.19. A lot of the training and backtesting as well as trading + ranking aspect is well documented in README.md for those interested in using the system for their own. We essentially use a ML technique called Ensemble Learning that uses agents. These agents range from simple strategies in TA-Lib to more proprietary agents (we plan to make this feature open source as well) that model trades done by each investment firms (as posted on marketbeat and changes in portfolio value on 13f reports). The ensemble learning part occurs behind the scene with each agent's parameters ((skew, flip ratio etc.) - there's about 82 parameters) being contorted in different ways in a controlled manner so that it's fine tuned with agents from same class being given feedback loop to their respective control files. This is done using 1m tick from Intrinio although we anticipate moving to Databento. The open source version is not the same as our proprietary one but it has the same framework (mostly because a lot of services we use are paid). We want our users to be able to use AmpyFin without having to pay a single cent. - Target Audience Institutional traders want to benchmark their trading AI agents against other publicly available agents without having to share their proprietary models, and retail investors want clear, AI-driven trading signals without analyzing complex strategies themselves, so, Ampyfin solves both problems by ranking multiple trading agents—including strategies, investment portfolios, and AI models—and assigning decision weights to generate the most optimal buy/sell signal for each ticker - Comparison There really isn't any application like this out there to be fair. A lot of trading systems utilize one complex strategy and still use human traders. Signals are there for the human traders. In terms of for retail investors, a lot of application require private information to access their data. We don't. We don't require any personal information to use our application. - The Team To be quite frank, we are currently a small team spread out in different locations. We're all software engineers full time. We mostly work on the project Friday evening - Sunday evening. There's no set amount of time one needs to work. The team is just there so that our efforts are united in pushing out certain features by a certain flexible timeframe while grabbing a pint. We all stand by the same goal for the project which is keeping and keeping the project open-source, providing full transparency to our users, and maintaining it so that it is free to use. In fact, currently, we're using Open AI's Pro+ for the LLM side to scrape data from Benzinga and Marketbeat but we're implementing an effective web scrapping tool for out users to use in the open source version.