Skip to main content

AI Builder

Overview​

This section outlines the structure and functionality of the AI Builder system, comprising two primary components: the Requirement Gatherer and the Component Builders (which further include the Flow Builder and Action Builder).

1. Requirement Gatherer​

The Requirement Gatherer is a conversational, streaming AI agent designed to interact with end-users and elicit detailed requirements through a structured conversation. Its key functionalities include:

Conversational Interaction​

  • Engages with users by asking targeted questions to clarify and refine their requirements.

Structured Output​

  • Translates conversational data into a structured component draft, including initial properties such as:
    • Name
    • Purpose
    • Input/Output
    • Initial README

Thread Management​

  • Maintains state using conversation threads, enabling users to have multiple concurrent conversations without data loss or confusion.

Workflow​

  1. User interaction via a streaming conversational interface.
  2. Requirement extraction into a structured draft format.
  3. Passes an initial, undetermined component draft to the Flow Builder.

2. Component Builders​

The Component Builders comprise two distinct but interrelated components:

2.1 Flow Builder​

The Flow Builder receives component drafts from the Requirement Gatherer and recursively evaluates each draft for scope and clarity. It primarily focuses on organizing components effectively by determining whether a component needs to be split into subcomponents.

Features​

Recursive Component Evaluation​
  • Assesses each component draft for clarity, scope, and separation of concerns.
  • Creates new subcomponents as needed, building a hierarchical component tree.
Component Categorization​
  • If a component contains subcomponents, it is classified as a Flow.
  • If it contains no subcomponents, it is classified as an Action (draft).
Event Streaming​
  • Progress and decisions are streamed via Server-Sent Events (SSE) for enhanced introspection and real-time monitoring.

Workflow​

  1. Receives a structured component draft.
  2. Recursively evaluates and breaks down the component into a tree structure.
  3. Streams progress and categorizes components into Flows or Actions.
  4. Returns a final event containing a list of components (one or several action drafts and potentially one or several complete flows).

2.2 Action Builder​

The Action Builder focuses on implementing detailed action components by writing and refining Python code.

Features​

Python Code Generation​
  • Generates robust, context-aware Python code based on the provided action component draft.
Optional Enhancements​
  • Research: Performs web-based research related to user requirements prior to coding.
  • Static Analysis: Conducts linting and type checking to ensure code quality.
  • Code Review: Reviews final code for alignment with requirements, security best practices, and overall code quality.
Sandboxed Environment​
  • Handles package management, environment variables, code editing, and testing within an isolated sandbox.
Event Streaming​
  • Progress updates (tools used, reasoning, coding status) are streamed via SSE for real-time insight.

Workflow​

  1. Receives an action component draft.
  2. Optionally conducts preliminary web research based on requirements.
  3. Generates initial Python code.
  4. Optionally performs static code analysis and improvements.
  5. Runs comprehensive testing until satisfactory results are achieved.
  6. Optionally performs a detailed review of the code.
  7. Streams the entire process for visibility and debugging.
  8. Returns a final event containing a complete action component with fully implemented code.

System Diagram​

Summary​

The AI Builder provides a robust, scalable solution for translating user requirements into actionable, well-structured software components. Its modular design promotes clarity, scalability, and maintainability, supported by real-time streaming capabilities for transparency and user engagement.