Love your work! Would you consider creating a detailed video on deploying RAGs using FastAPI at a production level? Specifically, I’d love to see how to maintain conversation IDs across multiple sessions, manage state for multiple chains, and set up a scalable infrastructure for concurrent user interactions.
That´s more something for my capstone project of my Udemy course. For YT these very advanced and complex topics are most of the time not worth the effort (unfortunately)
@@codingcrashcourses8533 Thanks for the reply, totally understandable. Looking forward to the Udemy course and appreciate the great content you share on UA-cam!
First of all I would like to thank you personally for this video! Such an amazing and quality content again!! It gave me the basis for many ideas to implement!! 😊 I would like your opinion, if having both vector tables and simple ones in the same db could be a good practice for a chatbot project based on the one in the Azure LangChain course. Thank you again for your help!! 😊
love your work, really amazing, but i have a question hope you can reply or make video, so with langgraph, can we make a graph have 2 nodes father and son , both or 2 different chatbots, the entry is the father i can keep talking to him when ask him about talking to his son i will transition to the son chatbot (i made till here) but i want to keep talking to the son until he says goodbye so it moves to _end_ is that even possible (i just simplified it by father and son but the real use case are 2 agents and i can keep talking to 1 of them until he transition me to the next agent so i keep talking to him,...)
Ok, so what you want is some determinism. That´s indeed not so easy. I would try use a classifier in the parent, which decides to route to the child. You save that output in a state variable and use that value to bypass the parent and directly talk to the child agent. But the child agent needs a classifier too, where it decides weather to continue the conversation or not. I would do create a TypedDict like this class SharedState(TypedDict): current_agent: str You start with "parent" as value, then you overwrite it in the parent class with a classifier. You then bypass it based on the new value. And the child agent resets current_agent to "parent". Does that help? :)
yes that's so helpful , i tried it out it kinda worked but always pass by the father, what i want is like "time travel" this a powerful hard architecture especially if there's a lot of agents, kindly if you make a video on it would be so helpful for the community ❤️🤘
I was referring to the idea of directly transitioning between different chatbot agents without reverting back to the parent each time. Essentially, after switching to the 'child' agent, I want to remain in that context until the conversation with the child agent naturally concludes, without the need to route back through the parent repeatedly. This behavior mimics a more seamless, 'time-travel-like' jump where once we move to the child agent, we stay there until an explicit handover or exit point is reached. This kind of flow ensures a more dynamic and contextually persistent experience, avoiding the overhead of constant re-routing and maintaining a clear, continuous dialogue state.
@@immortalx678 I think I understand what you want, but I think you need to still use the router. But only the first time it´s LLM based, the second time you use the variable to route. I see no way around that tbh. You need the information of what subagent you talk to
Love your work! Would you consider creating a detailed video on deploying RAGs using FastAPI at a production level? Specifically, I’d love to see how to maintain conversation IDs across multiple sessions, manage state for multiple chains, and set up a scalable infrastructure for concurrent user interactions.
That´s more something for my capstone project of my Udemy course. For YT these very advanced and complex topics are most of the time not worth the effort (unfortunately)
@@codingcrashcourses8533 Thanks for the reply, totally understandable. Looking forward to the Udemy course and appreciate the great content you share on UA-cam!
You’re the best for so many reasons, a great creator and teacher, have learned a lot from you.
Nice timing of the video. Lots of usecases are trying to implement SQL agent
First of all I would like to thank you personally for this video! Such an amazing and quality content again!! It gave me the basis for many ideas to implement!! 😊 I would like your opinion, if having both vector tables and simple ones in the same db could be a good practice for a chatbot project based on the one in the Azure LangChain course. Thank you again for your help!! 😊
sorry for the late response - I think this is totally fine and even the preferred way of doing it.
im loving your videos, it is realy helping me. When you are going to release the langgraph course?
To be honest, I am behind schedule. Got so much other stuff to do currently :(
Hi Markus, Is your Udemy LangGraph course still online? I cannot seem to find it.
@@DiegoMolinaingmecanico Hello Diego! I still work on it, sorry but it takes more time than I expected
@@codingcrashcourses8533 Thank you fo the prompt response, I understand. Looking forward to getting on to it when is available!
@DiegoMolinaingmecanico i will make a Promotion Video ok it. Just activate the bell :)
love your work, really amazing, but i have a question hope you can reply or make video, so with langgraph, can we make a graph have 2 nodes father and son , both or 2 different chatbots, the entry is the father i can keep talking to him when ask him about talking to his son i will transition to the son chatbot (i made till here) but i want to keep talking to the son until he says goodbye so it moves to _end_ is that even possible (i just simplified it by father and son but the real use case are 2 agents and i can keep talking to 1 of them until he transition me to the next agent so i keep talking to him,...)
Ok, so what you want is some determinism. That´s indeed not so easy. I would try use a classifier in the parent, which decides to route to the child. You save that output in a state variable and use that value to bypass the parent and directly talk to the child agent. But the child agent needs a classifier too, where it decides weather to continue the conversation or not. I would do create a TypedDict like this
class SharedState(TypedDict):
current_agent: str
You start with "parent" as value, then you overwrite it in the parent class with a classifier. You then bypass it based on the new value. And the child agent resets current_agent to "parent".
Does that help? :)
yes that's so helpful , i tried it out it kinda worked but always pass by the father, what i want is like "time travel" this a powerful hard architecture especially if there's a lot of agents, kindly if you make a video on it would be so helpful for the community ❤️🤘
@@immortalx678 What you want the time travel for?
I was referring to the idea of directly transitioning between different chatbot agents without reverting back to the parent each time. Essentially, after switching to the 'child' agent, I want to remain in that context until the conversation with the child agent naturally concludes, without the need to route back through the parent repeatedly. This behavior mimics a more seamless, 'time-travel-like' jump where once we move to the child agent, we stay there until an explicit handover or exit point is reached. This kind of flow ensures a more dynamic and contextually persistent experience, avoiding the overhead of constant re-routing and maintaining a clear, continuous dialogue state.
@@immortalx678 I think I understand what you want, but I think you need to still use the router. But only the first time it´s LLM based, the second time you use the variable to route. I see no way around that tbh. You need the information of what subagent you talk to