I have an exciting challenge for you: build a web app using a RAG (Retriever-Augmented Generation). The app should have a split-screen interface; the left side is for chatting, and the right side displays the original document. As users ask questions, the responses should include clickable links. These links, when clicked, will automatically scroll to and highlight the specific part of the document used to generate the answer. This feature is particularly useful for scrutinizing critical information in legal or medical documents, as it allows users to verify the responses and read more directly from the source. It took me a week to build this, and it turned out to be awesome. Pain points: coordinates , JavaScript, marking pdf, Hint: Metadata in your embeddings.
@@dharniesh3651I ended up using Aspose because I couldn’t get the coordinates to match using the python library and the JavaScript libraries. I did create my own markdown that worked, but I was told I couldn’t modify the original document in any way. :-(
great video, keep up the good work bro. do it also do auto syncing option so if data is updated in source then it can auto update its base context ? as i am building for production ?
Thanks for you video, that is great, Will you please explain how to get this "headers = { "Accept": "text/event-stream" }"? What if the url is on the company's site after login? Can I put the authrozation in the headers as well to use the Reader API? Please advise
I have an exciting challenge for you: build a web app using a RAG (Retriever-Augmented Generation). The app should have a split-screen interface; the left side is for chatting, and the right side displays the original document. As users ask questions, the responses should include clickable links. These links, when clicked, will automatically scroll to and highlight the specific part of the document used to generate the answer. This feature is particularly useful for scrutinizing critical information in legal or medical documents, as it allows users to verify the responses and read more directly from the source. It took me a week to build this, and it turned out to be awesome.
Pain points: coordinates , JavaScript, marking pdf,
Hint: Metadata in your embeddings.
Hey, if you are making it open source pls give the GitHub link here
That's a great idea!. I wonder what pdf reader are you using to match the coordinates ?
@@dharniesh3651I ended up using Aspose because I couldn’t get the coordinates to match using the python library and the JavaScript libraries. I did create my own markdown that worked, but I was told I couldn’t modify the original document in any way. :-(
@@dharniesh3651 I just switch to using AsPose. So much easier and it works great. They have both, a python library and JavaScript among others.
great video, keep up the good work bro. do it also do auto syncing option so if data is updated in source then it can auto update its base context ? as i am building for production ?
Very Nice!!
Does it also has the functionality to parse the data from PDF?
Awesome bro. Thank you for posting such great content.
Thank you sir
thank, for all, good job
Thank you sir
is there any way to use this on websites that you need to login to?
Can you give me an example?
@@raunaqss pretty much scraping any website that needs login. Would be more for personal projects nothing commercial
wow this is helpful
Thank you sir
uau thanks!
If I passed a certain website and that website has several sub-domains, does it go through the subdomains as well?
May not
The answer is NO. You can try it on their website using the Try Demo option
Thanks for you video, that is great, Will you please explain how to get this "headers = {
"Accept": "text/event-stream"
}"?
What if the url is on the company's site after login? Can I put the authrozation in the headers as well to use the Reader API?
Please advise