Rajib Deb

Пікірлер

@Gamezone-kq5sx 23 сағат бұрын

Very informative

@AlexCasimirF 7 күн бұрын

Thank you a lot for this video! There is quite some content about Mem0 on KZbin already - but as it is with most hyped or promising frameworks, most creators unfortunately do not take the effort of actually looking at the codebase and explaining how it works, but rather just scrape the surface by showing off the usage examples they found in the docs in the "quickstart" section. Contributions like yours are what open source actually needs, so thanks again!

@seadude 7 күн бұрын

I’m curious, how much better are the results than if you just issue a prompt to summarize directly to GOT-4/o/o1? The results would need to be objectively better in order to justify the agent complexity.

@rajibdeb4059 7 күн бұрын

Agree, this is something you may need to experiment and see if this approach gives better result especially when you have large transcripts with domain nuance. In that case you may have to have a fine tuned model for the hints

@shaiknaveed78 8 күн бұрын

Thank you Sir for the video, but how is this different from adding few shot examples to the prompt. Do the few shot examples we pass in a prompt are same as the hints here? Please help understand, Thanks again!

@rajibdeb4059 7 күн бұрын

Yes, if you can add the hints manually for every use case, it should work like the hints. This approach will be more automated

@IdPreferNot1 12 күн бұрын

Nice, Speed scroll capture the images and then send the images in one parralllel batch inference and restitch for faster speed. Gemini flash 1.5 is multimodal and allows 15 free rpm at <128K token window and can batch process images easily withing this limit.

@darkmatter9583 14 күн бұрын

@darkmatter9583 15 күн бұрын

Hi can you help to learn?(Not laws i mean your other content) I would like to reach your level and learn from you please, algorithms but also application side like karparthy,fe fei li to be able to opensource and be able to do something (like healthcare, entertainment,finances)etc

@short.positive849 16 күн бұрын

What's the difference between doing this and trying to call a db to save the response to it.

@seththunder2077 23 күн бұрын

Also, will u provide github link to this project?

@rajibdeb4059 23 күн бұрын

@@seththunder2077 it is there

@seththunder2077 23 күн бұрын

This is amazing. Is there a way we can let the LLM to also display the graphs or results on the streamlit app because thats an important feature

@yazanrisheh5127 22 күн бұрын

This would be a cool update to have

@pythonlover20 27 күн бұрын

To deploy, you need can check out the integrations section.

@yazanrisheh5127 Ай бұрын

Would love more videos on multion. Do you think it can be used in production/

@igris8569 Ай бұрын

Hey, is the hub depreciated now ? what handles that now ? for integrating into langchain

@rajibdeb4059 Ай бұрын

It is part of langsmith

@yazanrisheh5127 Ай бұрын

How is this different from AgentOps?

@OscarDiaz-fb8jd Ай бұрын

This is the best training on this topic so far

@seththunder2077 Ай бұрын

Could you do integrating it with langchain please

@rajibdeb4059 Ай бұрын

please take a look here github.com/rajib76/memory_examples/blob/main/examples/01_remember_user_preference.py

@user-wr4yl7tx3w Ай бұрын

does relying on events to trigger sufficient to cover all use cases?

@rajibdeb4059 Ай бұрын

@@user-wr4yl7tx3w some use case may need choreography some orchestration. So both patterns need to be considered. This is no new concept, microservices 101

@afrasiyab95050 Ай бұрын

How much effort was it to set something up like this? Is this scalable for more complex data?

@NikhilThorat-x7n Ай бұрын

Hi Sir - Cannot access ur git link, please can you keep the project open.

@coladeu. Ай бұрын

the number of open tabs made me understand memory at a different level

@kitranet Ай бұрын

Does thie studio only works in Machine?

@rajibdeb4059 Ай бұрын

@@kitranet I am using it through langsmith. You need to have at least plus membership of langsmith. There is a desktop version of studio which works on certain Mac machines only

@BharatVarsh47 Ай бұрын

Is there git code for this?

@HumanDefenseForce Ай бұрын

Thanks for the mem0 video

@pariajeihooni5150 2 ай бұрын

@ajibdeb4059 Thanks for the video. I didn't find any content related to connecting agents to the knowledge base. Could you help with that?

@nikcool8614 2 ай бұрын

Hello ... nice tutorial video ... What is the datatype of embedding in the code in python ?

@princemathew9034 2 ай бұрын

Hi Rajib, I am in a dilemma, New to Langchain agent framwork. What i like is the abstractions and length of features it provides to build a true agent. What i fear is that it is too abstract and there are icebergs underneath which i dont have any clue. What if something goes wrong? Difficult to troubleshoot and may be i do not have the flexibilty enough to quickly do a work around. So i am thinking of devloping a very basic functionality which i will have full control. Any thoughts or your experience of taking Langchain to production? Your feedback is appreciated. THanks

@AyushKumarSingh-l1p 2 ай бұрын

Thanks for the video, loved how you went through every single step!

@MrEravila 2 ай бұрын

Its easy to see that the physics teacher agent was built after the math agent since the description in the physics teacher agent still references to mathematics and the instructions in the moderator related to the physics teacher agent mention maths not physics. The typical error of a copy & paste.

@IdPreferNot1 2 ай бұрын

Nice video. Would like to see the next step for expanded Supervisor Agent capability. I appreciate that you are doing this without abstraction as i want to understand this before looking at a langgraph version, but would like to see that too :)

@LoganMarkewich-o9w 2 ай бұрын

There is an orchestrator, but it could be exactly an event-driven orchestrator :) Great video!

@rockypunk91 2 ай бұрын

Do we need to reindex all documents, everytime we add new document. Is there any way to run it programitically

@dikshyakasaju7541 2 ай бұрын

Amazing explanation!! But if we don't change the existing prompts then, there's no point in creating 'Parser Lambda functions', right?

@OluebubeOnwuamaegbu 2 ай бұрын

what about the prompt templates? Does this video only cover handling the response in the lambda function? What exactly are the prompt templates? How are those passed to the LLM?

@reverse_3258 2 ай бұрын

Hi @rajibdeb4059, @EccleezyAvicii, do you know how to implement partial response? like if the Tool takes time to provide response, I will show temporary message to the user? Your prompt reply will be greatly appreciated.

@OluebubeOnwuamaegbu 2 ай бұрын

I don't think I could've found a better video. As an intern I am so grateful thank you so much !

@dikshyakasaju7541 2 ай бұрын

Great Explanation! However if we have to build a KB for prod, would it not be better to create a separate lambda function where we can have more control over how we create KB using Langchain and upload it to Vector DB? And then create a seprate RAG Agent with Lambda Function and Open API?

@dikshyakasaju7541 2 ай бұрын

Can't we create a KB using LangChain because that way we have more control over how the chunking for the documents are done?

@krunaltanna8469 2 ай бұрын

Good explanation Rajib, May I know how is it different from when we create a KG using Neo4j?

@GirishPatil-k9r 2 ай бұрын

Which Authentication type you set for your tool ?

@munkuo5 2 ай бұрын

You can decompose the compound questions into simple questions. The Langchain page has a good example for this. Thanks for this video!

@AaronRicharteGranell 2 ай бұрын

Can u provide the code that goes in the tool schema?

@dana1virtual 2 ай бұрын

thank you! very clear approach and code!😶

@munkuo5 3 ай бұрын

Can you share the colab / github link? Awesome work, been following you for some months now.

@tara1473 3 ай бұрын

Hi Rajib, great work. What should we do if we want to create index and use Vertex AI Matching Engine to index the multimodal embeddings ( including for example text and image embedding for each item)? how should this be done? Does matchingengine index support that?

@mohamedbassiony9322 2 ай бұрын

Have you gotten the answer?