WEBVTT

00:00.360 --> 00:04.080
Hey there, Eden here and I hope you enjoyed the project so far.

00:04.520 --> 00:12.120
And in this video we want to be adding end to end self rack and self rack is derived from the self rack

00:12.120 --> 00:19.720
paper and is basically meaning that we are going to reflect on the answer that the model generated.

00:20.040 --> 00:25.560
So we are going to take the generation and we're going to compare it with the documents, and we're

00:25.560 --> 00:28.240
going to see whether the model hallucinated or not.

00:28.240 --> 00:33.360
So we need to check whether the answer is actually grounded in the documents.

00:33.560 --> 00:35.680
Now, if it's grounded, then that's cool.

00:35.680 --> 00:37.400
And we're ready to the second step.

00:37.400 --> 00:40.160
And the second step is to reflect on the answer.

00:40.160 --> 00:46.520
However, now we want to reflect whether the answer answers the question that the user originally asked.

00:46.520 --> 00:52.760
And if the answer is yes for that, then this means we can return the answer to the user, and if the

00:52.760 --> 00:58.640
answer does not answer the question, then this probably means that we need to make a web search, because

00:59.000 --> 01:02.560
we're not going to find any additional information in the vector store.

01:03.000 --> 01:10.920
And if we hallucinated the answer so it's not grounded in the documents, then we want to regenerate

01:10.920 --> 01:14.560
it again and to ground in the documents.

01:14.560 --> 01:15.200
The answer.

01:16.080 --> 01:19.000
So that's what we're going to do in this video.

01:19.160 --> 01:21.200
We're going to implement it end to end.

01:21.240 --> 01:22.480
We're going to write the chains.

01:22.480 --> 01:23.600
We're going to write the tests.

01:23.600 --> 01:24.800
We're going to add the nodes.

01:24.800 --> 01:30.040
And we're going to add, of course, all of those conditional branches and see you in the code.