Chat-driven Generation

User Problem

Users currently face a steep learning curve and lack of flexibility when creating applications and components like charts and tables using the platform's existing specification-based approach.

• Difficulty with Format: It's often unclear what format or structure the prompt should follow, leading to overwhelming complexity and trial-and-error.

• Slow Iteration: The process is slow, taking around 30 seconds for each chart to generate. This forces users to craft comprehensive, detailed initial instructions instead of allowing for quick, fluid, and iterative adjustments.

This leads to a less intuitive experience where users spend too much time defining requirements upfront instead of getting instant visual feedback.

What is it?

The new Chat-driven Generation feature provides an intuitive, conversational interface for building data applications and components. It uses a Large Language Model (LLM) within a chat environment to interpret user instructions, dynamically generating the component, and creating a specification to accompany it. It shifts the creation process from a rigid, spec-based method to a guided, conversational workflow, without losing the benefits of a spec for judging a component’s accuracy and sharing with colleagues.

What does it allow users to do?

Chat-driven Generation empowers users with a more natural and efficient way to build. It allows users to:

• Iterate Fluidly: Users can issue individual, rapid instructions to quickly adjust or refine a chart, guiding the LLM toward the final result in a dynamic, step-by-step conversation rather than waiting for long initial builds.

• Attach Context Easily: The chat environment naturally retains the conversation history, allowing users to effortlessly attach context from previous instructions to their current request.

• Inspect and Save Specifications: While the user interacts with the chat, the system automatically generates an underlying specification (the code for the component). Users can inspect this specification for accuracy, save it permanently, and share it with colleagues to ensure reproducibility and collaboration.

• Use Generated Specs as Context: Saved specifications can be used as the starting point or context for future chart creation, streamlining repetitive tasks or building variations on existing designs.