Create Bedrock Agent:
Step 6:
Deploy Bedrock Agent:
Step 3:
Inference parameters:
Temperature: 0
Randomness
Degree of variability in response
no variability
Top P: 1
most probable tokens to generate next
Don’t want creativity
Top K: 0
choose smaller number of next output text
Don’t want creativity
Responsible AI:
Guardrail: harmful categories & prompt attacks filters:
Step 1:
Step 2:
Step 3:
Guardrail: denied topic:
Name:
Medical Advice
Definition For Topic:
Discussions that involve providing guidance, recommendations, or suggestions related to health, medical treatments, symptoms, or overall wellness.
Sample Phrases:
- How to manage chronic pain?
- Recommend a diet for heart health.
- What to do for persistent headaches?
- Is it safe to combine medications?
- How to improve mental well-being?
Step 3:
Step 4:
Guardrail: sensitive information filters:
Name: Account Number
Regex pattern: [A-Za-z]{6}\d{4}
Guardrail behavior: Mask
Add description - optional: Matches account numbers in the format XXXXXX1234
Guardrail: contextual grounding check:
Step 1:
Step 2:
Attach Guardrail to Bedrock Agent
Step 1:
Step 2:
Advantages of AWS Bedrock Agent:
Secure Data:
- Data securely executed on AWS user VPC.
Seamless Integration:
- Integrates seamlessly with the SDK.
Auto Scalable Operations:
- Auto scales to handle data volumes.
Reduced Operational Overhead:
- Minimizes on-premises infrastructure.
Simplified Compliance:
- Facilitates compliance with data regulations.
Flexible Configuration:
- Offers customizable configuration options.
Top comments (0)