Step 1: Create a Project

Sign up at Vellum and create a project for your AI application. Organize your prompts, test cases, and workflows in one place so your whole team stays aligned.

Step 2: Design Your First Prompt

Use the prompt builder to write instructions with variables for dynamic input. Vellum lets you test multiple prompts and model configurations side by side before you commit.

Step 3: Create Test Cases

Add example inputs and expected outputs to a test case suite. These become your safety net, letting you verify that changes to your prompt or model don’t degrade quality.

Step 4: Run Evaluations

Execute your test suite and review the evaluation results. Vellum scores outputs so you can compare versions objectively and pick the winner with data, not guesswork.

Step 5: Build Multi-Step Workflows

For more complex AI features, use the workflow builder to chain steps like classification, retrieval, and generation. This lets you create agents and assistants without writing extensive glue code.

Step 6: Deploy and Monitor

When you are ready, deploy with the API or SDK. Vellum tracks usage, cost, and performance so you can monitor your app in production and iterate confidently.

Final Thoughts

With Vellum, building production-grade LLM apps is methodical and measurable. Design prompts, test rigorously, deploy with confidence, and your AI features will be ready for real users.