Skip to content

architecture.md

Latest commit

 

History

History
65 lines (52 loc) · 3.73 KB

architecture.md

File metadata and controls

65 lines (52 loc) · 3.73 KB
Raw

System Component to Code Module and LLM Utility Mapping

UI Layer

  • Gradio UI: Handles user input, diagram type selection, preview, and download.
    • Code: GradioInterface (to be implemented in the main app, e.g., UMLBot_streamlit_app.md or a new UI module)
    • Interacts with: ChatBotService (service layer)

Service Layer

  • ChatBotService: Orchestrates the flow from user request to diagram generation.
    • Code: New module/class, e.g., UMLBot/service/chatbot_service.py
    • Uses: NLPParser, DiagramBuilder
  • NLPParser: Converts free text to UML constructs using LLM.
    • Code: New module/class, e.g., UMLBot/service/nlp_parser.py
    • Leverages: llm_utils/aiweb_common/generate/ChatServicer.py, PromptAssembler.py for LLM calls and prompt construction.
  • DiagramBuilder: Builds PlantUML code from parsed model, manages validation and repair.
    • Code: New module/class, e.g., UMLBot/service/diagram_builder.py
    • Uses: PlantUMLValidator, ThemeManager, PlantUmlClient
  • PlantUMLValidator & AutoRepairStrategy: Validate and repair PlantUML code.
    • Code: New module/class, e.g., UMLBot/service/plantuml_validator.py
    • Can leverage: LLM utilities for advanced repair (e.g., ChatServicer for LLM-based repair suggestions)
  • ThemeManager: Supplies PlantUML themes.
    • Code: New module/class, e.g., UMLBot/service/theme_manager.py

Infrastructure Layer

  • PlantUmlClient: Renders PlantUML code (remote or local).
    • Code: New module/class, e.g., UMLBot/infrastructure/plantuml_client.py
    • Implements: HTTP and local JAR rendering, with failover logic.
  • Database, Auth, Monitoring: Standard integrations for user/session/history, authentication, and logging.
    • Code: To be implemented as needed, e.g., UMLBot/infrastructure/db.py, auth.py, monitoring.py

Supporting Services

  • DownloadController: Serves images to the UI.
    • Code: Part of UI or a small API endpoint.

Chat-Based UML Revision Workflow

  • The chat workflow enables users to iteratively describe, revise, and correct UML diagrams through conversational interaction.
  • The UI (Gradio/Streamlit) relays user messages to the service layer, which orchestrates prompt construction, LLM invocation, and diagram updates.
  • Each chat turn can request new features, corrections, or explanations; the system updates the PlantUML code and diagram accordingly.
  • The workflow supports error correction by leveraging LLM-based auto-repair and surfacing actionable feedback to the user.

Generic Error Handler Integration

  • The generic error handler is integrated across all layers:
    • UI: Displays status and error messages, provides fallback diagrams.
    • Service: Captures errors from LLM, prompt, and rendering steps; attempts auto-correction.
    • Infrastructure: Handles network and rendering failures, reporting issues to the service layer.
  • All errors are routed through the error handler, ensuring transparency and robust recovery.

LLM Utility Integration

  • All LLM calls (NLP parsing, auto-repair, chat-based revision, error handling) should use the llm_utils/aiweb_common/generate/ classes:
    • ChatServicer for conversational LLM calls.
    • PromptAssembler for prompt construction.
    • QueryInterface for abstraction.
    • ObjectFactory for extensibility (e.g., plugin new diagram types).
    • WorkflowHandler for orchestrating multi-step flows and logging.
    • GenericErrorHandler for capturing, reporting, and auto-correcting errors.

This mapping ensures that all LLM-related logic and error handling are centralized and reusable, following the project's coding standards and leveraging the common codebase.

Next steps: Specify in detail how and where to leverage these LLM interface classes and error handler in the design.