I have just found the best rag system out there that is completely open source. Introducing Graph Rag by Microsoft AI. Graph Frag is a project that is a data pipeline and a transformation suite that is designed to extract meaningful structured data from unstructured text using the power of different large language models. Now, before I even go forward and explain what Graph Rag is, I need you to understand what RAG is in general. For the people who know RAG, basically stands for Retrieval-Augmented Generation natural language cering approach. It is something that is used to enhance existing large nage models, and this is with external knowledge so that the answers to questions are more relevant in case they require specific knowledge. This is why RAG is something that many of us tend to use. It's an approach that is the cheapest and the most standard way to actually enhance a large nage model with additional knowledge for the purposes of answering a question. In addition to this, it's also shown to reduce the tendencies of large ST model for hallucination so that the generation adheres more to the

information from the context which is generally more reliable, and this is something that GraphRAG does the best at the moment. Now, besides question-answering, RAG can also be used for many different things with natural language processing tasks, such as information extraction from text, recommendations, sentiment analysis, summarization, and this is all within a private local storage. So now that we know what GraphRAG is, or RAG in general, what is the difference between GraphRAG versus traditional baseline RAG systems? Well, GraphRAG is something that is combining graphs with retrieval-augmented generation. It combines text extractions as well as network analysis and LLM prompting and summarization into one powerful system. Unlike simple text search methods, GraphRAG is something that uses knowledge graphs to extract and organize information that's going to improve the accuracy and the relevance of responses, especially for complex or private datasets. It's something that's going to connect disparate pieces of information and provide synthesized insights that are going to be outperforming baseline RAG techniques. It's something that is following a

holistic approach that's going to enhance the understanding and the summarization of large data collections, and it's going to make graph probably the best choice for leveraging large stage models in advanced data analysis. Is in question answering tasks, which is why I definitely see this as the best ultimate framework for rag. Now before we get started, I'd like to introduce world of AI solutions. This is a really big update that has been launched for my channel, and this is where I have compiled team of software engineers. We have machine learning experts, AI consultants, and this is basically a team where we're going to be providing AI solutions for businesses, as well as personal use cases where AI solutions can be implemented to automate certain things or to help business operations. Now if you're interested, take a look at the ti form link in the description below. Now before we get started, I definitely recommend that you take a look at the patreon page so that you can access the new subscriptions that will be releasing this week. If you would like to book a consultant call with me, you

can do so with the link in the description below as well so throughout today's video. I'm going to be showcasing how you can get started with graph frag as well as basically understanding some of the capabilities of now. I really want you guys to watch this demo video, which is going to explain graph rag a little bit further in detail by the creators of this project. So take a look at this video, and we'll be right back. We are excited to introduce graph rag, a new approach to retrieval augmented generation. Graph rag sends entire private data sets through a large language model to extract details and construct a knowledge graph that represents this content. These knowledge graphs are then used to perform retrieval augmented generation at query time. Let's start by taking a look at one of these graphs that we've created for this demonstration. We've indexed all of the transcripts for Kevin Scott's behind the Tech podcast. As the transcripts are ingested, we ask GPT 4 Turbo to extract all entities and relationships after parsing graph rag.

created the following Knowledge Graph to orient you. Each node represents an entity, and the size of each node is relative to the number of nodes it is connected to. For example, Kevin Scott is shown here as the prominent node that is connected to most everything. This makes sense as he is a host of the show. Each color represents a topic of interest. These colors show the semantic communities that GraphRAG builds using graph machine learning during the source data indexing. For this data set, most topics are centered around the specific episode and guest, which we can see in this example that I'm showing here. Using GraphRAG, we have shown dramatic improvements for several types of questions where RAG is known to fail. While typical vector search, RAG is excellent at performing retrieval for highly specific facts, it lacks the ability to perform any operations that require higher-order reasoning and understanding over the data set. For example, let's consider the query, What are the most odd conversations discussed? Traditional vector search will find similar text to the question and provide that back to the LLM to answer.

question with this Vector similarity approach works well if the query provides specific content that can be closely matched with the source data. What are the most odd conversations discussed doesn't match any specific content and instead requires a holistic understanding of all the content. This is where graph frag can help. While building the knowledge graph, we aggregate semantic concepts and derive holistic understanding of the whole basket of sources. When we query graph rag, the knowledge graph guides our content discovery to provide the llm diverse, comprehensive, and topically relevant content at the right level of abstraction. So let's get started and start installing the prerequisites that are essential for this to be functional on your computer. First things first, you're going to need to make sure that you have your programming language, which is Python, so install the latest version. Once you have done that, you need to make sure that you have Pip install, and you can easily do so by just simply copying this. Go into your command prompt, simply just open that up by searching up your command prompt on Windows, and just paste

this in and click enter. I'm not going to be doing this since I already have it, but this is to install the package. You're going to need to clone the repository. So what I am going to be doing afterwards is going into VS Code, which I definitely recommend that you install. This is going to be the ID that I'm going to be using to install this application. And in this case, what you would want to do is just simply open up a new window on VS Code. You want to click on the toggle panel button to open up your terminal. Once you have done that, you want to simply go into bash and you want to basically type in PIP install rafre. So once you have cloned your repository as well as installed the packages by doing the PIP install command, what you can do now is type in CD graph rag. Once you have done that, you're going to need to export your API key. So to do so, you can just simply type in export graph rag underscore API key and then you can just simply paste

in your API key from OpenAI. I'm going to be showcasing how you can switch other large language models around, even utilize AMA, which is like local models for this. So, in that case, I'm going to start off by putting in my API key from OpenAI, and then I'll be showcasing what you can do next. And simply after you paste it in your API key, you can just simply click enter, and you should be good. Now, what you're going to need to do is create an input folder for allowing all your files or documents within that input folder. So, to do so with in bash, you want to type in mkd, and you want to put input. And the reason why I'm naming it input is because we're going to be placing in all of our files within that category or that folder. So, to access these folders, what you can do is you can just simply go on click on file you want to then click on open folder, and you want to basically open up the folder as to where you Clon this repository, and you can see

that our input folder has been created. So, for this example, I'm going to be taking a Ford financial report for this year, and I'm going to be placing it into this framework to get the best rag strategy or system implemented within this report, so that I can get the best answers and I can basically get the best analysis for financial reports for this particular case. So, I have gone along and I basically just inputted the text output from this PDF within VS Code. Now, I know this was kind of inefficient when I could have just placed in the PDF itself, but you can obviously paste in different file types within this. In this case, I just pasted in a txt, and we're going to be moving on with this example now. So, in this next step, what we're going to be doing is pasting in this command: python DM rrag index. And basically, this command is to index the current document that we have so that we can start implementing this rag system to it, meaning that we're going to be able to now chat with this.

document so you're going to be able to do this every time you are chatting with a new file. So you would want to index this, and to access this, you can see that it has now have a community report on. Basically, the report that we have created, it focused on indexing almost every bit of that particular for document that we had uploaded. Now, if you are to go over to the EnV file, you're can paste in your API key from OpenAI. But say if you're going to be basically working on utilizing another model, you can configure this over here by specifying the type of model. You can specify if you're going to be utilizing an OLAMA model, or you can basically just utilize it within the OLAMA interface, which I truly recommend. You would want to simply paste in the endpoint. You can do this with Gro and so many other models. So, this is where you would want to configure this. After you have made the changes, you can just simply click on file and then save, and you can start running it from there. In essence, this

yam file is going to be utilized to configure the chunks. The input which we have set already and have it so that all of our files will be set. Here you can change the cache, the storage, change the reporting format. You have different entities. You can change markdown and so many other things within this uh y file. And that's basically the configurations of what sort of output you want with this rag system. And lastly, we're going to be running the code and that's by just pasting in this command. So we can just paste this in, which I'll leave a link to in the description below so that you can easily paste this command in. And we're going to click enter. Within a couple seconds, it's going to now focus on running this code. Now after it has finished indexing, you're going to be able to see the output folder being created. And now you can basically start chatting with this. And to do so, you just simply paste in the same command, python space m-m graph rag, but this time we're going.

to paste in query, and this is basically to initialize the query or the input. And once you have done that, you can then have it so that it's set to the root folder. And then you want to click space and then Dot. And after you have pasted in this command that we used previously, what you will need to do afterwards is put another space. And then you want to keep it method Global by simply just pasting in this command. And you want to put another space and put your query in. And that's basically how you can implement the rag system within your documents. This is just a way for you to access it locally. I definitely think and recommend that you utilize it with AMA, which is quite easy to set up. You just need to set your endpoints. But that's basically for today's video on ra rag. I definitely see this as the best Ultimate rag system that's out there, and it's something that is completely open source. I'll leave a link to whatever I use in today's video in the description.

below make sure you follow me on the patreon. Make sure you follow me on Twitter so that you can stay up to date with the latest AI news. And lastly, make sure you guys subscribe, turn on notification Bell, like this video, and check out our previous videos so you can stay up to date with the latest AI news. But with that thought, guys, have an amazing day, spread positivity, and I'll see you guys fairly shortly. Peace out, fellas.

GraphRAG: Ultimate RAG Engine - Semantic Search, Embeddings, Vector Search, & More!

World of Ai is here to push the creative use of Artificial Intelligence (AI) applications for simple day-to-day tasks!

Here at WOA, you will find tips, tricks, and guides using AI applications to help you out in your day-to-day life. Going to strive to make your life easier! I will show you how and what you need to do to help your life out! I will strive to show you the wonders of AI and if that interests you, then make sure you subscribe to the channel now!
