SyGra Studio adds a visual builder for synthetic data workflows in SyGra 2.0.0

Synthetic dataset generation is moving from config files to visual tooling, and ServiceNow is pushing that shift with SyGra Studio—an interactive UI in SyGra 2.0.0 aimed at making LLM-driven pipelines easier to build, inspect, and rerun with governance in mind.

Visual workflow design for synthetic data generation and prompt engineering

SyGra Studio turns a SyGra task into a canvas: teams can assemble nodes (for example, LLM steps for drafting and summarizing) without hand-editing YAML. For marketers and e-commerce operators, the immediate win is iteration speed—prompt engineering becomes less about switching between terminals and more about tweaking prompts with inline variable hints pulled from the dataset schema.

Studio supports guided model setup across common providers and runtimes (including OpenAI, Azure OpenAI, Ollama, Vertex, Bedrock, and vLLM), then lets users connect data sources like Hugging Face datasets or local files and preview rows before committing a run. Column names from the preview become state variables you can reference directly inside prompts, which reduces wiring errors when multiple steps depend on shared context.

Execution observability, debugging, and exportable configs for automation teams

Where Studio targets B2B operators is observability. Runs stream node-by-node status, latency, token usage, and estimated cost, with execution history written to local artifacts (for example, under .executions/). There are also developer-focused controls—inline logs, breakpoints, and a Monaco-based editor—so you can debug failures without leaving the UI.

Importantly, Studio doesn’t replace the underlying framework: it generates the same SyGra-compatible graph config and task executor scripts you would commit in a repo. That means teams can prototype in UI, then operationalize the output in CI or other automation contexts.

For a concrete reference, ServiceNow points to an existing workflow that consumes the Glaive Code Assistant dataset from Hugging Face and iterates answers until a critique loop returns “NO MORE FEEDBACK,” using a conditional edge to route revisions or exit cleanly. Dataset reference: glaiveai/glaive-code-assistant-v2. A walkthrough and release context are available via SyGra 2.0.0.

The bigger signal: as AI teams standardize synthetic data as a repeatable asset, tooling that combines previews, versionable configs, and run-level metrics is becoming table stakes—not a nice-to-have.