Synopsys R&D · Agent systems
Building memory and reliability into intelligent systems.
Reduced memory-step latency by 67% (12s to 4s) by parallelising three query rephrasers with conditional result merging. Validated agent-memory accuracy across approximately 8,000 test cases by integrating graph-based memory retrieval and self-contained query rephrasing. Improved multi-step agent reliability with dependency-aware sequential execution for code generation and execution workflows, and built a corrective query tool for targeted recovery. Collaborated with development and testing teams to integrate new Ansys Mechanical APIs, resolve test issues, and integrate Copilot capabilities into the Scripting Agent.
Outcome
Reduced memory-step latency by 67% (12s to 4s).
Evaluation
Validated agent-memory accuracy across approximately 8,000 test cases and used testing collaboration to resolve issues while integrating new APIs and Copilot capabilities.
Focus
Worked on agent memory, graph-based retrieval, self-contained query rephrasing, dependency-aware execution, corrective recovery, and integrations around the Scripting Agent.
Contribution
Worked on agent memory, graph-based retrieval, self-contained query rephrasing, dependency-aware execution, corrective recovery, and integrations around the Scripting Agent.
Engineering decisions
Parallelized three query rephrasers with conditional result merging · Combined graph retrieval with self-contained rephrasing · Added targeted corrective querying for recovery
Evaluation
Validated agent-memory accuracy across approximately 8,000 test cases and used testing collaboration to resolve issues while integrating new APIs and Copilot capabilities.
Public boundary
This case study intentionally excludes proprietary code, screenshots, internal architecture, and unreleased product details.