finrobot

An open-source AI agent platform designed to enhance financial analysis through specialized agents powered by Large Language Models (LLMs).

FinRobot

Welcome to Quant Lab. Quant Lab in some of my previous videos I've talked about Finn GPT and FinRobot. These two were projects from a, you know, foundation called AI for Finance and it's a nonprofit organization which is dedicated to advancing AI in the finance industry. 

So they have three large projects. One is called Fin GPT which I've talked about and was very foundational in my own understanding of how, you know, AI and AI related areas can be plugged and joined with finance. And then reinforcement learning framework which is called Fin RL which is the third one in the framework. 

Now both of these were a little tech heavy and, uh, honestly the barrier to entry was high specifically, uh, limiting some of the abilities or some of the capabilities. So in this video what I want to do is hop on to Fin Robot which is a very, very interesting framework and platform. 

So stick around the video till the end. You'll gain a lot of understanding about how AI can be plugged into the, you know, domain of financial engineering. So let's dive in. 

All right, so this is the website GitHub whatever you want to call it FinRobot and it's an open-source AI agent platform for financial applications using large language model. Kind of mouthful but wordwise that what it is meaning or referring to. 

All right before we go there if you're new to my channel I talk a lot about technology related for trading be it core fundamental concepts or be it Python coding integration with brokers coding out strategies. And finally I also talk a lot about AI and, you know, adjoining areas just to give you a share AI Agent Store with friends.

Perspective of so if you have not subscribed please subscribe. Uh hit comment hit like. Share this video with other people so that it reaches the right set of people. It will help me out in achieving my goals of democratizing knowledge around the space. 

Part two of the announcement I am going to start a batch pretty soon for fundamentals of python for financial engineering. It's a good exhaustive course meant for absolute beginners. Basic eight-hour course or like 50% of Q&A so completely rich from a doubt perspective and you'll gain a lot from it if you new to the space. If you want to build strategic knowledge if your intent is to learn it for quick learning strategies so that might not be uh for you. But do check out the link it is in the description and uh stay tuned for it. 

I might organize an ad hoc webinar for it uh you know for 2 hours on this Sunday so stay tuned to the Community page and uh I'll keep posting. Let's come back to fin robot right. Uh the idea is is financial robots and what are robots they are automations that can help you achieve some piece of information automated way. Open source code is available readily and AI agent platform inter. What is an AI agent right this is an image from a back.

A not or anything but use this attribution so I hope the guys please read out the article. It's actually good enough very good article as well. 

So AI agent usually we talk to large language model like chat GP type answer. It's a very one-to-one interaction. 

Second kind of interaction can be you give it some information about the operation or the data or the domain you have. Let's say you upload some PDFs and you ask question based on that. Or you bring in your organization data and ask questions on that. Or you bring in some external data and you ask questions on that. That is what is called as rag architecture retrieval augmented generation LM. 

And why it does not know is because it's trained on a certain point in time. Like its knowledge base is not like absolutely latest or doesn't have the knowledge which is specific to your company your needs. Like chat GPT might not know what your company is doing. So you can augment that information. 

So that's point number two. Agents, you know agents take this to multitude of different scenarios and use cases. Agent, it's a think of it as a u entity which understands natural language and it can generate task list. It understands natural language, understands prompts, generate task lists and execute those operations as well. 

Get me the stock price of or forecast of, you know, a company in India. So it needs to understand and break the problem into get me the stock price of a company. So first it will need to do let's say, you know, data access by API fin data. Second, it will understand the modeling paradigm forecast predict.

Third it will use some techniques to generate forecast and finally give me an answer to there can be a multi-step processes. So as you can see in this diagram screen U, you a user asks an AI agent a question and it one it can go to any data document. Two it can execute some form of code like it's a code execution API call and three it can connect to any kind of uh ml model. It can be actually any kind of model and you can further give it to a chat llm or an optimization problem or a prediction problem or a forecasting problem. 

Forecasting problem finance stock price forecast stop loss forecast optimization problem optimization will be you know I have this portfolio composition give me the right combination of it. Prediction could be you know predict the return books of yeah give me the balance sheet given balance sheets of a company for past 10 years predict the PE or something like that and there can be multitude of use cases. 

And then finally it can go to any llm any other llm that you want it to be from meta Google U etc etc. So so AI agent is a wild concept if you ask me interesting exciting but it is important to know this if you are into building these kind of systems and uh if you're trying to get into this space for uh using it for trading or stock market analysis or you're building platforms that can help other uh Traders as well. 

So so that's what I would say is a AI agent. Now coming back to AI for finance AI for finance is a framework or platform which brings all of this together in a single place and hence the framework and ecosystem exist. 

Please share the AI Agent Store with friends.

I'm going to talk about what happened. Why is it sorry about this? It should not, uh, I'll just image go. Okay, maybe this is just hosted somewhere where I cannot access it or there's some internet problem. I don't know. 

But so fin robot has there. There's different layers. So at the very base is the foundation models. This part clicking is a bad idea. This part multisource llm Foundation models, which are, you know, Lama, chat GPT, Falcon, any, uh, language model. Then you have another layer which is l M Ops and data Ops, which is, you know, LM Ops, large language model. Go there. Dat Ops is all data related, financial data, andl Vector databas, etc., etc. 

And on top of it, you can have a financial llm, something like Finn GPT or any other custom model which you have built. And on top of it, other agents. Now these agents could be, uh, having a variety of use cases. Let's say, for example, you have Market forecasting agents, which is trading strategist, multimodel agent like chart data. 

And I talked about how you can include chat, uh, chart like technical charts into, uh, chat GPT and it will give you an analysis. So things like that. And then there can be document analysis agents which could be doing risk analysis, earning analysis, Market analysis, unstructured data analysis. 

There could be a plethora of Agents. Something like I'm interested in scraping the data of, uh, sort of, let's say, certain news providers in the country. Let's say moneycontrol.com. Although it's illegal, but you're gathering data. Let's just say not scraping, you're gathering data, right? Uh, you can have an agent which gathers the data. 

Share the AI Agent Store with friends.

Every day build a sentiment and throw it out at you. So is po framework the whole framework is intended to be uh having a multiple layer four layers. I think the top layer is the agent AI layer. The second is the financial llm algorithm layer which are very tuned model for financial domain. Then it's the data and llm Ops layer and then multisource financial model. 

Now as a Trader how useful is this for you? This is completely useless that you have learned and saw the video. I think what you should be asking yourself is how is this going to impact me. So in the future if you are an analyst a lot of these jobs can be automated by this. If you are somebody who's building a company which Can U do market analysis and publish reports for your customers a lot of that can be automated using this. If you are somebody who does a lot of technical analysis for himself or for his customers and Publishers report frequently all of that can be automated using this. 

If you want to get into uh getting basic initial recommendations of algorithms and assessing how different algorithms are behaving like have the right pieces of code three four different algorithms back testing logic and have them evaluate you can have an agent for it. Essentially this technology has and I say that with all the humility I have and don't mean to cause any chaos but this technology has the potential to turn into something big couple of years down the line where it can happily replace uh you know let's say the low level or the inter the I would say beginner level jobs for sure. So yeah. 

Share the AI Agent Store with friends.

That's why I'm bringing these conversations out to you agent workflow. So there's two three different uh setups that has to be understood. 

So first is a user proxy user sense it to into a an action group which is let's say this. Then there is a different combinations of uh let's say agents that are existing. One is a director agent, other is an assistant agent, another is an llm analyst agent. 

Kind of roles and personas I'm looking in that screen. Sorry for not shifting eyes. So different kind of llms can there agents can there llm agent which is like a boss agent. It has different analysts doing different analysis and it consolidates all that. 

It's possible you can have an assistant which is you know two pair and you can have a it's it's all possible all within the realms of possible sub pass. Each of them have access to all kinds of data imaginable. They can look at market news, they can look at financial, they can look at trading records, they can look at market depth, they can look at volume, they can look at delivery reports. 

What else can they look? They can look at chart, they can calculate technical indicators, they can do textual representation etc etc. All of this is powered by the brain uh what they are calling as inner reflection. Let's not go there. I don't think the intent is to explain how it works but it's rather to explain. 

And then all of these can connect to different data source API, Yahoo Finance, Open BB, Fin LP. Sub say integrate you can have a multitude of tool calling interface which is also very much possible with this. 

So yeah, that's the agent. Share the AI Agent Store with friends.

Workflow I would say uh how important is this very very important. How scary it is very very scary. Uh this is a more like a scheduler conversation that they are trying to say ke how do you get uh the selection of each agent and llm for each task. That's a good uh thought process to work around but take I think it will be too much to get into. 

And then this article gets into all sorts of code and advanced configurations. I have not yet tried it but I would believe this will be really really uh fundamentally possible. Like forecaster are present technical writing like equity research reports are possible. Uh you can automate a lot of it. I'm trying I'm I I'm going to work up some of these and I know there's lot of projects that I have. 

But yeah if somebody's interested happy to partner with them. Uh and then there's bunch of references around AI agents masking data chat GPT etc etc. Now uh that's what I wanted to talk about in this video. Uh this became actually a longer video than I had originally thought. 

Uh so yeah I'll hope you have watched the video and enjoyed the video. Thank you guys for listening. Bye-bye.

Note, that there may be slight differences in text and video. Watch video for the original content.

FinRobot Video