Introduction to LangChain (Udemy.com)

Good course, I have learnt a lot. The only thing that’s missing…is the updated material. Since the course was released, langchain and even openai’s GPTs have changed a lot so I had to spend a lot of time figuring out how to overcome all these changes. But the course is quite good and Damien explains everything very nicely!

The examples in section 7 is not working. We are missing packages that had to be installed ie pillow-heif, pikepdf and i think unstructured-pytesseract We also had to brew install cmake, poppler, tesseract, poppler-utils and pkg-config. another problem now is that langchain kernel crash when doing this command: raw_pdf_elements = partition_pdf( filename=path + "llava.pdf", extract_images_in_pdf=True, infer_table_structure=True, chunking_strategy="by_title", max_characters=4000, new_after_n_chars=3800, combine_text_under_n_chars=2000, image_output_dir_path=path + "images/", verbose=True ) any suggestion how to solve would be nice.

The course is full of content and cover lots of material and great example of use cases, which I love! I learn a lot of LangChain, and other related libraries. I am not sure if it's LangChain itself, which I realised is a very messy library, or the instructor's style, but the more I progress into the course into the advanced lessons I understand less and less the material and I just follow recipes. Why sometimes I use RetrievalQA.from_chain_type(retriever=...) and then in another time we call RetrievalQA(retriever=...)? What does this FAISS actually do? Why do we sometimes use Prompt template and in other times Chat prompt templates or Message Prompt Templates? Why in one time he uses from langchain.docstore.document import Document and in another time he uses from langchain.document_loaders.base import Document (which fails for me)? What's the difference and why is it necessary? Each video takes up to 10 minutes, and shows the instructor running code, saying a sentence on each line on code such as "now I need to create a retriever so I need to run this.", but still I keep finding myself spending much more time on each video trying to figure out what the code actually does, which I thought should be explained in the videos. At the same time, given the amount of content, I can understand that there is a trade-off between quantity and depth..

This is a fairly new field, this course does a great job of providing context in an organized fashion without living too much in the theory. At times I'm left wishing for more context on each package and and the purpose, but then I am reminded the very tool we are learning is good at answering questions. ;)

As I continue to watch the course, I become more and more disappointed. 17 lectures so far and there are several running themes: a) Code examples come before talking about what's coming. So, we get caught up in code examples without knowing why we need to learn that. That might not be that bad if the instructor didn't overwrite code when he was making points. b) The lecture notes (written materials) come at the end of the lesson. But even if you download it earlier and follow along, you MUST pause and watch every step in the video while he is coding because the written material doesn't cover some of the code that he writes, executes, and overwrites. c) Some of the code that worked in his lectures, no longer work. And because we are not sure what the purpose of the code is, I had to spend time figuring out how get sample data into the format that his video/and written materials expected... (the weather_data example) just find out that his point was not about that dataset, but about what you can do with CSV datasets. So we could have easily used a different, easier dataset to try CSVLoader I will refine the score if it gets better throughout the course. I paid for it, so I will finish. He is explaining in a very good way so It's still a good deal. I am just frustrated at the organization. Tips: 1. Before starting the lecture, download the code and the lecture notes (Written material) 2. Copy the code from the written materials, copy and paste into your own notebook. 3. Copy the lecture transcript, paste it into your own jupyter notebook. 4. Then listen to the video and run the code when he gets there. Its longer, and harder this way, but it's probably the best way to get maximum learning of this course.

As a technical learner, it is appreciated that the content is concise and practical. Too much theoretical content obfuscates how to apply the technology in the real world. It is hence invaluable for the course to deliver in a manner where it forms the basic building block, a balance of theory and practicality, for the learner to extrapolate and build more advanced implementations.

Very nice course, it is very useful how Damien explains every step which make it more easy to understand. Multimodal RAG has been particularly challenging since I have a Windows PC not a Mac pro, but finally I manage to reproduce everything. This is great about this course you not only follow the instructions, you also need to resolve problems related to your specific context and environment.

This is easily one of the best courses I've encountered on Udemy. Damien's approach is incredibly comprehensive, covering every detail from setting up the environment to advanced concepts. The course is easy to follow and packed with information, making it an excellent journey from beginner to expert in a remarkably short time.

Written material in the end of each section is very helpful. We can keep that open while attending the session to get more understanding on the topics. Few sessions the content looks like too much and repeated what is there in the screen, so that can be reduced. Overall very informative course to understand end to end picture of LLM and LangChain with great examples.

So far it is a good course and clear. However there is not enough explanation of each function or imported function of packages used. The course mainly appears to be a "watch my copy paste code and run it" as a demonstration, rather than a learning activity or showing what they actually do, and what variables do.