WEBVTT

00:01.240 --> 00:02.960
Hello everyone and welcome.

00:03.600 --> 00:07.920
In this video I will walk you through how the course is organized.

00:08.720 --> 00:17.880
We will begin with LM fundamentals where we will learn about LM model architecture, its capabilities,

00:18.800 --> 00:23.360
and also expose the inherent vulnerabilities that it has.

00:24.640 --> 00:28.240
Then we would move on to AI guardrails.

00:29.160 --> 00:33.800
Why is there a need for AI guardrails?

00:34.320 --> 00:41.920
We would expose some of the security risks and business justification for implementing protective measures.

00:44.000 --> 00:52.400
Then on we would cover some of the core techniques like vector embeddings and retrieval augmented generation

00:52.400 --> 00:53.160
concepts.

00:55.400 --> 01:03.300
Then we would introduce you to the user input guardrails where we would use a couple of models like

01:03.660 --> 01:12.700
Prompt Guard and Llama guard from meta, which are open source models to detect input detection and

01:12.700 --> 01:13.860
content moderation.

01:16.140 --> 01:19.740
Then we would introduce you to the foundation model guardrails.

01:20.220 --> 01:30.100
We would use Pi three hallucination judge and evaluation models for detecting output quality.

01:33.220 --> 01:41.260
Then we would move on to understanding Garrick, which is an open source vulnerability scanner that

01:41.260 --> 01:47.060
provides standardized detection for comprehensive LLM security testing.

01:48.740 --> 01:53.900
Then we would dive deeper into cybersecurity using AI agents.

01:55.100 --> 02:01.120
We would create autonomous software programs that would help us detect.

02:03.760 --> 02:07.840
The vulnerabilities and help the.

02:10.920 --> 02:14.200
Help the system with better decision making.

02:15.360 --> 02:23.000
And then we would have evaluation frameworks where we would use different metrics driven assessments

02:23.360 --> 02:32.680
for hallucination detection and context relevancy, followed by AWS bedrock integration, where we would

02:32.680 --> 02:39.480
use the bedrock implementation for AI guardrails.

02:40.320 --> 02:48.080
And then we would use couple of open source tools like guardrails, AI, and guardrails framework for

02:48.280 --> 02:52.160
implementing custom security implementations.

02:54.080 --> 02:56.840
Thank you everyone and hope you enjoyed the course.