Abdalrahman AttyaGrounded support · Serverless AWS
Customer Support Assistant
A focused generative-AI workflow in which customers receive policy-grounded answers, review a proposed escalation, and explicitly confirm before a support ticket is created. Authenticated staff then advance tickets through resolution.
- Release
- v1.0.0
- Evidence
- Local, CI, browser, and AWS lifecycle verified
- Boundary
- Verified environment removed after acceptance
The workflow
A customer asks a policy question. The service retrieves relevant passages, applies deterministic and managed safety controls, and uses Amazon Bedrock to compose a cited answer.
If human help is needed, the assistant summarizes the issue and asks for confirmation. Idempotency prevents a retry from creating a second ticket. Cognito identity and operator-group authorization protect the staff queue.
Architecture
Grounded responses with controlled escalation
Engineering decisions
- Grounding: answers use the supplied fictional policy source and retain citations.
- Confirmation: model output cannot silently create a persistent support ticket.
- Separation: customer sessions and staff permissions use distinct controls.
- Cost-aware operation: expiry, throttles, retention, alarms, and exact-target teardown constrain the environment.