Built on large language models and an industry knowledge base, the agent generates test scripts and control code in multiple languages, with intelligent completion, comment generation and error parsing. Script development is 3-5x faster.
Script development is 3-5x faster, code acceptance reaches 70%, and repetitive coding is reduced.
Supports mainstream languages including Python, Java, XML, C and JS, covering testing, control and diagnostics.
Completes code automatically, adds standard comments, parses error messages and suggests debugging steps.
Deployed at an NEV manufacturer with incremental pre-training, SFT and RAG retrieval, covering 6 large projects.
A large language model serves as the reasoning core, tuned on automotive R&D and testing corpora to support requirement parsing, root cause reasoning and code generation.
Brings together ZD simulation knowledge, historical fault cases and diagnostic experience with retrieval-augmented generation, so output matches real engineering contexts.
Supports local and private-cloud deployment so R&D data never leaves the company, meeting OEM data security and compliance requirements.
Agent output connects directly to test execution, simulation scheduling and ticketing systems, so results flow automatically and are tracked to closure.
Around 70% code acceptance (31% used as-is plus 41% with minor edits), more than 40,000 lines generated across 6 large projects, script writing nearly twice as fast, used by more than 400 engineers.
Three-stage tuning with incremental pre-training, SFT and RAG retrieval, covering test scripts, diagnostic scripts and PLC and fixture control code.
ZD Technology · AI agents for vehicle R&D and testing