WEBVTT

00:01.080 --> 00:02.600
Hey there, Ethan here.

00:02.600 --> 00:09.520
And in the next couple of videos, we'll be implementing Chirag or correctly react based on the correct

00:09.760 --> 00:11.200
rank research paper.

00:11.560 --> 00:18.280
And this is an advanced rank technique that is going to help us get more quality answers when performing

00:18.280 --> 00:20.200
retrieval, augmentation, generation.

00:20.880 --> 00:24.080
And the basic concept of Seerat is pretty straightforward.

00:24.280 --> 00:29.960
We first want to start by taking our query, performing in the vector search, semantic search, and

00:29.960 --> 00:32.680
retrieving relevant documents from our vector store.

00:33.200 --> 00:38.840
After we have all those documents, we want to start and self-reflect, to critique those documents

00:38.840 --> 00:44.440
and to determine whether they are indeed relevant to our original query or not.

00:44.880 --> 00:51.320
If all of the documents are relevant to our query, then this is a happy flow and we simply want to

00:51.360 --> 00:56.280
augment our original prompt and send everything to the LLM like we always do in RAC.

00:56.680 --> 01:02.400
However, if you find document that we find out that are not relevant to our query, of course we want

01:02.400 --> 01:08.080
to filter the null and we also want to perform some external search on the internet and get to more

01:08.080 --> 01:09.120
information.

01:09.120 --> 01:13.800
And to augment our prompt with that real time information we get from online.

01:14.440 --> 01:19.320
And after we have that, we can augment our prompt and send everything to the LN.

01:19.640 --> 01:23.880
And this technique is going to give us much more quality response.