Quality assurance · Production

Amazon AMET Payments team reduces QA test case generation from 1 week to hours with SAARAM multi-agent AI

The problem

The AMET Payments QA team manually analyzed BRDs, design documents, UI mocks, and historical test preparations for each new feature, consuming 1 week per project and requiring one full-time engineer annually just for test case creation.

First attempt

Initial single-agent AI attempts fed entire BRDs to one agent and produced generic, non-actionable test outputs due to context length restrictions, lack of specialized processing phases, and hallucinations creating irrelevant test scenarios.

Workflow diagram · grounded in source
1
Document routing at entry point
routing
“The file type decision agent serves as the system's entry point and router. Processing documentation files, Figma designs, and code repositories, it categorizes and directs data to appropriate downstream agents.”
2
Parallel specialized data extraction
ai_action
“The Data Extractor agent employs six specialized subagents, each focused on specific extraction domains. This parallel processing approach facilitates thorough coverage while maintaining practical speed.”
3
Multi-type Mermaid visualization
ai_action
“the Visualizer agent, and it transforms extracted data into six distinct Mermaid diagram types, each serving specific analytical purposes. Entity relation diagrams map data relationships and structures, and flow diagrams visualize proces…”
4
Data condensation and synthesis
ai_action
“The Data Condenser agent, and it performs crucial synthesis through intelligent context distillation, making sure each downstream agent receives exactly the information needed for its specialized task. This agent, powered by its condense…”
5
Multi-subagent test generation
ai_action
“The Test Generator agent brings together the collected, visualized, and condensed information to produce comprehensive test suites. Working with six Mermaid diagrams plus condensed information from Agent 4, this agent employs a pipeline …”
6
Test suite deduplication and delivery
output
“the Test Suite Organizer performs deduplication and optimization, delivering a final test suite that balances comprehensiveness with efficiency”
Reported outcome

SAARAM reduced test case generation from 1 week to hours, cut QA validation effort from 1.0 FTE to 0.2 FTE, identified 40% more edge cases than the manual process, and achieved 100% adherence to test case standards.

Reported metrics
Test case generation time1 week to mere hours
QA engineer effort1.0 FTE to 0.2 FTE
Edge cases identified40% more
Adherence to test case standards100%
Show all 5 reported metrics
test case generation time1 week to mere hours
QA engineer effort1.0 FTE to 0.2 FTE
edge cases identified40% more
adherence to test case standards100%
payment success rateIncreased through comprehensive edge case testing
Reported stack
Amazon BedrockClaude SonnetStrands Agents SDKPydanticAmazon Q Developer for CLIAmazon Bedrock AgentCoreAmazon CloudWatchMermaid
Source
https://aws.amazon.com/blogs/machine-learning/how-the-amazon-amet-payments-team-accelerates-test-case-generation-with-strands-agents?tag=soumet-20
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

SAARAM reduced test case generation from 1 week to hours, cut QA validation effort from 1.0 FTE to 0.2 FTE, identified 40% more edge cases than the manual process, and achieved 100% adherence to test case standards.

What tools did this team use?

Amazon Bedrock, Claude Sonnet, Strands Agents SDK, Pydantic, Amazon Q Developer for CLI, Amazon Bedrock AgentCore, Amazon CloudWatch, Mermaid.

What results were reported?

Test case generation time: 1 week to mere hours; QA engineer effort: 1.0 FTE to 0.2 FTE; Edge cases identified: 40% more; Adherence to test case standards: 100% (source-reported, not independently verified).

What failed first in this deployment?

Initial single-agent AI attempts fed entire BRDs to one agent and produced generic, non-actionable test outputs due to context length restrictions, lack of specialized processing phases, and hallucinations creating ir…

How is this quality assurance AI workflow structured?

Document routing at entry point → Parallel specialized data extraction → Multi-type Mermaid visualization → Data condensation and synthesis → Multi-subagent test generation → Test suite deduplication and delivery.

Related quality assurance cases