I have been doing dozens of trainings and courses in the past months and I have to say this was the absolute best deep dive into RAG!!! I loved how u showed what this looks like in code. Most explanations are completely theoretical. This was amazing! please continue to post videos on (gen)AI!!!! I would personally would love to understand how a system (CRM, etc.) is integrated in this architecture. Anybody else interested in this? Please like!
I’m a cybersecurity professional but I don’t code. More like an infrastructure guy. But now I’ve become very interested in AI security. Coding was always daunting for me. This guy made it look easy as pie. Thank you!
It's so good to see a good demo, please put more videos, I would specifically looking for RDMS database to vector database to store embeddings so my user can use AI
I had been looking for the full set steps for a year now. This is easiest and very good video content! Thank you for teaching such important use of LLMs!
I tried the approach with a product catalog dataset and was able to get similar results. You are a wonderful teacher . Thank you so much for this video and for the education. Looking forward to the next demo.
Amazing job Don! Subscribed as soon as I have seen your first video about RAG and this one is even greater. Thank you very much, please keep making videos. You have a talent for teaching to lay audience.
Many thanks, Don. Your explanation was simple, clear and very helpful. To ensure that the top 5 answers are more relevant to the prompt questions, how can we verify and validate the model's output ? How can we test the bias in the data being prepared using the CSV, which is necessary to obtain the top answers? My other question is, how can we integrate these models in real-time? i.e. how to integrate these models into different HIS and CIS touchpoints (or any healthcare system, really), how to run and train these models, what additional computation (hardware requirements and increase in total cost of ownership) might be needed, and how to confirm User Acceptance Tests and seek customer sign-off before moving to production.
In terms of validation, you will need to do some of this by hand - meaning a human reading the answers and verifying that they are accurate, that they represent the right tone, and that your model isn't responding to things that are off topic and/or are dangerous. These 3 could be separate validation threads.
Thanks for the video! It’s great to see how the RAG really works. I understand that RAG is quite good at organizing answers with text. How about replacing the CSV files with large Excel sheets and needing answers from different columns in the same row? Will the RAG perform well in that case? I find that while the response is often correct, sometimes it jumps to different rows and makes up answers. What is your take on using Excel sheets as documents for RAG?
Great explanation! Is there a way i can get the code you show in the jupyter notebook please? Would like to play around with it using my dataset. That would be very helpful!
Excellent Don, the video explained RAG as neatly as possible! 1. Is it possible to share the code and the web source and web data csv file so we can follow along for practice? 2. Then openAI model used here for embedding, did you have to pay for using that? 3. Could you cover a separate video on current AI challenges related to AI ethics and ways to tackle them? Eg: AI used to create spam/ fake profiles, generate non consensual adult content images etc.
Thank you, this was one of the most useful hands on tutorials and really cleared the confusion. I wonder if you plan to share the Jupyter notebook you used and if there is an upcoming (or existing) video that address the step you took to gather the input data (i.e. scraping the website to create the csv). It would be so useful to know about techniques for data retrieval and storage from bigger sites/ other data sources, and what would be different if you didn't have the liberty to only pick the top 4 closest embeddings. E.g. what if it were a retail web site and the user asked "what dresses do you have in blue?". Thank you once again.
I've had a few questions on the 'scraping the website' step. I just basically wrote a python script, using a few libraries, that navigated through the website from the top, and stored the documents.
I really fall in love with RAG, I am a Master student and I want to do some research on RAG for my upcoming Thesis Project. Can I have some suggestion from you ? articles or publications ? Thanks for this video
Thanks for this. How would this differ if it were not web based public data, but actually patient data for example? How is access to permissions based sensitive data handled in this case?
Hi Don, I am learning that now and also doing something with assistant based on given material, like healthcare website in your case. So watched this RAG video to learn the concept. The question is that really needed to do the embedding and calculate the vector distances as opposed to just uploading all content (website text) to the storage and let the LLM to use it through a prompt. Means in this case the engine should basically do the same distance search, maybe cutting the material to smaller pieces than uploaded (?) and finally stick to the least distant for the answer. Do you think your more manual approach can generate better answers?
My question is, say you want to make a chatbot that gives responses to questions about workflow and how to do certain things within a company. Would it be possible to connect it to a database that holds lots of documentation about a company?
Yes certainly. With the approach I outlined, you can connect to documents within a database. You would just need to store the embeddings into the database itself, or a separate database that supports that.
Suppose I wanted the answer to cite its source(s) -- in your example, the web page(s) that went into the final answer. How would that change this approach? Presumably, I wouldn't want to encode the web page's URL in the same vector as the text as that would skew the similarity calculation (dot product).
Yes - you could put a reference in each of the documents that you choose and ask the LLM for the references back for the sections that it used. Given their are cryptic (like you could replace the URLs with unique IDs), they wouldn't really skew the embeddings.
The primary data structure used in data science is called a Data Frame. It's just like a SQL table. The rows are typically instances and the columns are typically attributes/fields/properties of each instance. Hence the variable df.
I have been doing dozens of trainings and courses in the past months and I have to say this was the absolute best deep dive into RAG!!! I loved how u showed what this looks like in code. Most explanations are completely theoretical.
This was amazing! please continue to post videos on (gen)AI!!!!
I would personally would love to understand how a system (CRM, etc.) is integrated in this architecture. Anybody else interested in this? Please like!
Awesome Don, please do more of these videos explaining more about AI/ML
next topic: one on Lang chain or llama model 3 which was released recently.
Yes - I'll take a look.
@@dwoodlock Thank you dan you are amazing
You really are a teacher, made it consumable to RAG newbies, this is how videos should be lol. Thank you very much
I’m a cybersecurity professional but I don’t code. More like an infrastructure guy. But now I’ve become very interested in AI security. Coding was always daunting for me. This guy made it look easy as pie. Thank you!
You are just great, as you know your subject and you have the will to elucidate us. Many thanks and all the best!
This is so clear! I am a beginner who was able to learn the whole process with your video, thank you so much!
This was one of the best videos on RAG. Very nicely done!
It's so good to see a good demo, please put more videos, I would specifically looking for RDMS database to vector database to store embeddings so my user can use AI
Try InterSystems IRIS.
The way you conveyed that information was so structured and informative.
Your explanation made me grasp the concept quickly.
Thank You!!!
I had been looking for the full set steps for a year now. This is easiest and very good video content! Thank you for teaching such important use of LLMs!
I am full stack developer trying to get into machine learning /ai industry workspace. Your videos are motivation for people like us.
I loved this explanation of RAG. I’ll share this with my team.
Best explanation I have seen to date of how RAG works.
I tried the approach with a product catalog dataset and was able to get similar results. You are a wonderful teacher . Thank you so much for this video and for the education. Looking forward to the next demo.
How did you set up. I tried to do what he showed in the video but cannot get past initial errors. For example- OpenAI not found
what website you used and how did you download the website? any suggestions?
Unbelievable content. Couldn't have been any simpler than this. Thank you so much.
Amazing job Don! Subscribed as soon as I have seen your first video about RAG and this one is even greater. Thank you very much, please keep making videos. You have a talent for teaching to lay audience.
This is awesome! Loved the explanation and walkthrough. Definitely sharing this with all folks i come across who'd need help with this!
Great video! Could you please do one video on RAG with knowledge graph? Thanks!
Thanks, after watching your video I got clear understanding how we can implement RAG.
Very thankful Don. It would be awesome to get some timestamps throughout the video, but this is incredible as is.
Ah. Good idea.
Done!
@@dwoodlock Awesome
I appreciate you immensely! Awesome concise clear information. I really hope you continue to make available content like this.
You made my demo remarkable, I explained my audience easily ❤❤❤❤❤❤
Fantastic stuff here.I hope you continue to explain the complex topics.
Great video! where can I get the code used in the video?
Many thanks, Don. Your explanation was simple, clear and very helpful.
To ensure that the top 5 answers are more relevant to the prompt questions, how can we verify and validate the model's output ? How can we test the bias in the data being prepared using the CSV, which is necessary to obtain the top answers?
My other question is, how can we integrate these models in real-time? i.e. how to integrate these models into different HIS and CIS touchpoints (or any healthcare system, really), how to run and train these models, what additional computation (hardware requirements and increase in total cost of ownership) might be needed, and how to confirm User Acceptance Tests and seek customer sign-off before moving to production.
In terms of validation, you will need to do some of this by hand - meaning a human reading the answers and verifying that they are accurate, that they represent the right tone, and that your model isn't responding to things that are off topic and/or are dangerous. These 3 could be separate validation threads.
@@dwoodlock Thanks Don and very helpful.
Always wanted to understand how everyone all of a sudden made their own bots. Now I understand how simple it is.
Great demo and explanation . Thank you for sharing the knowledge !
Amazing video, save me ton of days to study. Thank you very much Mr. Don
beautiful detaining. thankyou for a kind video
Appreciate you taking the time to get into the weeds
Thanks for the video! It’s great to see how the RAG really works. I understand that RAG is quite good at organizing answers with text. How about replacing the CSV files with large Excel sheets and needing answers from different columns in the same row? Will the RAG perform well in that case? I find that while the response is often correct, sometimes it jumps to different rows and makes up answers. What is your take on using Excel sheets as documents for RAG?
Good content. Thank you. Looking forward to next video.
Great explanation! Is there a way i can get the code you show in the jupyter notebook please? Would like to play around with it using my dataset. That would be very helpful!
Were you able to find the code shown in Jupyter or something similar?
Very useful! Amazing explanation!
Excellent Don, the video explained RAG as neatly as possible!
1. Is it possible to share the code and the web source and web data csv file so we can follow along for practice?
2. Then openAI model used here for embedding, did you have to pay for using that?
3. Could you cover a separate video on current AI challenges related to AI ethics and ways to tackle them? Eg: AI used to create spam/ fake profiles, generate non consensual adult content images etc.
I have the same questions. Did you find resolution
@@sandyna1k 1. You can try web scraping or sites like httracker.
2. Looks like it's a paid version of LLM.
3. Haven't found an answer yet.
Excellent presentation, well done. The code really helped.
Hi Don, thank you for the great video. It would be great if you could share the notebook codes 😊
Amazing Don. I wonder why is there a need to use Langchain. Can you please throw some light?
Great video, easy to follow, highly informative. Thanks!
Amazing video! Thanks Don.
Great video. Could you explain when would we use RAGs? Is it supposed to be used almost always with an LLM? Or RAGs have special use cases?
Thank you, this was one of the most useful hands on tutorials and really cleared the confusion. I wonder if you plan to share the Jupyter notebook you used and if there is an upcoming (or existing) video that address the step you took to gather the input data (i.e. scraping the website to create the csv). It would be so useful to know about techniques for data retrieval and storage from bigger sites/ other data sources, and what would be different if you didn't have the liberty to only pick the top 4 closest embeddings. E.g. what if it were a retail web site and the user asked "what dresses do you have in blue?". Thank you once again.
I've had a few questions on the 'scraping the website' step. I just basically wrote a python script, using a few libraries, that navigated through the website from the top, and stored the documents.
Thank you
Thanks for the video. Is there a way I can understand how the costing works when Im using RAG?
I really fall in love with RAG, I am a Master student and I want to do some research on RAG for my upcoming Thesis Project. Can I have some suggestion from you ? articles or publications ?
Thanks for this video
Great video very well explained. Do you also have a video on how to download a website into a CSV?
Not yet. I may do one - lots of you have asked.
What a wonderful example! Before I run through this from my own Jupyter Notebook - do you recall roughly how much it cost in OpenAI credits?
Thanks for this. How would this differ if it were not web based public data, but actually patient data for example? How is access to permissions based sensitive data handled in this case?
How did you turn the website into a csv file? web-scraper? if so, which one?
I wrote it myself using the beautiful soup library in python. It wasn't too complicated.
Great video! How did you get the content of the website? I am working on a similar project, and most of my effort seems to be scraping pages.
Hi Don,
I am learning that now and also doing something with assistant based on given material, like healthcare website in your case. So watched this RAG video to learn the concept.
The question is that really needed to do the embedding and calculate the vector distances as opposed to just uploading all content (website text) to the storage and let the LLM to use it through a prompt. Means in this case the engine should basically do the same distance search, maybe cutting the material to smaller pieces than uploaded (?) and finally stick to the least distant for the answer. Do you think your more manual approach can generate better answers?
Really good. Thanks!
Thank you Thank you!!!!
This was fantastic! Thank you!
Very Helpful, Can you Please make a video on RAG for low resource language like Nepali?
awesome explanation
can you share the notebook and the dataset?
love it, thank you! I appreciate these videos.
Thanks!
My question is, say you want to make a chatbot that gives responses to questions about workflow and how to do certain things within a company. Would it be possible to connect it to a database that holds lots of documentation about a company?
Yes certainly. With the approach I outlined, you can connect to documents within a database. You would just need to store the embeddings into the database itself, or a separate database that supports that.
Greate Explanation
Excellent, made so simple.😃
Simple and clear! Thanks!
Great Job, Thanks!!!!
helpful! thanks.
how do you get the webpage dataset?
I wrote a little python script that scraped the website by navigating it from the top like a tree-based search and stored the text of each page.
can you please share link to get the csv file, so that i can practice setting up this.
Suppose I wanted the answer to cite its source(s) -- in your example, the web page(s) that went into the final answer. How would that change this approach? Presumably, I wouldn't want to encode the web page's URL in the same vector as the text as that would skew the similarity calculation (dot product).
Yes - you could put a reference in each of the documents that you choose and ask the LLM for the references back for the sections that it used. Given their are cryptic (like you could replace the URLs with unique IDs), they wouldn't really skew the embeddings.
@@dwoodlock That makes sense, especially if I use IDs rather than URLs like you're suggesting. Thank you!
this is my first Vedio on AI, just trying this sample code, it keep asking API key ? how can i manage it?
Wow, Thank You
Thanks, this is so cool
Can you please share the csv file so that i can practice. I am a new researcher will learn these for my thesis.
You are great
How can we use a csv file with multiple fields as the prompt may require to evaluate all fields and answer the best one.
Most people will turn the multiple fields into a sentence or a paragraph. Then you have a document.
Awesome Sir !!
👏 new subscriber here
hi sir can i get this code to replicate on my jupyter notebook
is it possible to share CSV file Don?
For dummies please!!! step by step
So what is df?
Distance Formula I think
The primary data structure used in data science is called a Data Frame. It's just like a SQL table. The rows are typically instances and the columns are typically attributes/fields/properties of each instance. Hence the variable df.
That’s
What is concatenate?