Master LangChain #17 | LangChain ConversationEntityMemory | Entity Tracking#openai #genai

Key Takeaways

Welcome back to my uh text learning channel. I hope you all are doing really great and have watched my previous lang chain series. Uh if you haven't watched uh it is there in the playlist, you can go ahead and watch. Um and today's session is mainly on the another type of memory called the conversation entity memory uh which I'll be covering with the live coding examples how to use it. Um, so before we start, if you haven't subscribed to my channel, please go ahead and subscribe and give it a thumbs up. And if you have any specific topics which you want me to cover, please uh put it in the comment box. I can sh surely try to get you uh those at the earliest. Okay. So now let's see what is conversational entity memory, right? It is a specialized memory uh module that extracts the entities like name, the place, the object from your chat interactions, right? And tracks the um tracks the facts about them over time. It is very much important, right? It's specifically useful for AI agents that needs to refer back to previously mentioned entities meaningfully. So you can persist entities across sessions using customizable storage backends right. So basically extract entities it tracks the facts. It has the smart understanding which will be applicable for the agents and it uh is helpful for the agents to keep the the persistent uh of their history and you can also store those into your persistent storage and mainly useful for the chatbots. Okay. So now let's see what are the core features it offers. As you said, as I said, entity extractions identifies entities from the recent input and uh store it in the entities variables, right? And then you have entity summarizations. So summarizes what's known about each detected entity. Uh for example, if it is person, what is that person doing? Right? So that is there. Then entity store which uses this like a back end uh it could be um databases, radius, SQL to the pine cone for persistence. sync and as sync APIs are already part of it. So which we can start using it and also um buffer access like access chat history via buffer also is available and this is also depreciated but supported in the nchain 1.0.1. So now let's see how to uh start using this entity memory right. So let me create a file called uh wait a second this turn save go here and say conversation entity memory dot pi okay so before doing that I'll just take the base uh declaration or the modules or the the imports we need um I will will not need the say I I need the Azure chat open. I will need the uh summary memory also. We can take it right. Uh then we will have uh we will have uh uh we will have to import few other u uh modules also. So I'll input lang dot memory input conversation entity memory then I will conversation chain we already have it which is good the prompt template is already there right so these are the required ones now what I'll do is first I'll have to set the entity type entity template Okay. Uh because that is very much important to have the entity extracted. So what I'll do here? Okay. So what I'll do you are AI assistant um extract and uh remember key entities about the user. Okay. Then I'll say conversation history. Then I'll put history always I get spelling issue where say history then I will have to put known entities. So [clears throat] I'll just say entities this is where uh it will be already maintained right so that will be passed then I will say user message uh so which will be in okay then I'll say using the above information produce an appropriate response uh using the above information to provide and appropriate procreate response. Okay. So this is my template. Now I'll define the ROM template. Okay. Prompt uh wait prompt uh equal to ROM template from of I'll just give the entity. This is my prompt template. Okay. Now I'll define the memory. So I'll say memory equal to conversational entity memory of lm equal to lm. Then we will have uh return is equal to I'll make it true. Then I'll make conversation um conversation uh equal to conversation chain of llm [snorts] equal to lmm memory = then I'll have prompt= to prompt then we will have boo= Okay. Now what I'll do here is I'll just simulate it. Okay. So print of conversation dot predict of then I'll say input equal to hi I am Alex and I live in Paris. Okay. Then uh I will ask another question telling that what do you remember about me? What do you remember about me? Okay. Then also I will be printing the entities what it uh it is storing. Okay. So we have memory dot nt store. Okay, let's see if there is any syntax or the variable naming issue. Let's check it out. So we have conversation underscore entity think some issue with the ah yes spelling mistake is there so I'll have to convert sation Still the same conversation that's same right? Uh let me check it out. Um Okay, let's try from lamb chain memory put conversational entity memory. Let's see. Okay. I think below part uh yeah this is where the issues. Okay. Uh name prompt. Lots of [snorts] naming error. Okay, now I think syntax issues are gone. It started known entities. You see it has picked Alex and Paris. So if I say what do you know about me? I remember that your name is Alex and you live in Paris. And it has given the entity store with the summary. Alex is a person who lives in Paris. Paris is the city where Alex lives in. It has summarized and gave you the details. This is about how to use the uh entity summary, entity memory. Okay. Now let's see the overall summary. What we have seen it is entity extraction identifies entity names. Entity summarization holder summary for each entity. Then you can customize and store the data into the stores. Then you can support both async and sync operations. Then you have a buffer and it is also a one note that depreciation on the u this one right. So real world use cases is majorly character based dialogue like remember details like Alice works at this particular corporation right and customer assistance recall product names or account details across terms right that's very much important [snorts] and role play um AI like uh person details of a character location story right so in that case this is very much helpful so so now we have uh at the end of the session where we have covered what is conversational entity memory, how it helps, what are the real world use cases. If you have any specific questions, please put do put it in the comment box and please subscribe to my channel, give it a thumbs up and stay tuned for further series.