Graphs & Flowcharts

Graphs & Flowcharts Integration

The Graphs & Flowcharts module enables your AI agents to interpret, analyze, and reason across visual logic. By converting spatial diagrams and structural charts into machine-readable formats, Deeptrain bridges the gap between visual documentation and LLM reasoning.

Capabilities

• Structural Analysis: Identifies nodes, edges, and decision paths within flowcharts.
• Relational Mapping: Translates complex graph hierarchies into JSON or textual logic.
• Multi-format Support: Process logic from static images (PNG/JPG), vector graphics (SVG), or script-based diagrams (Mermaid/PlantUML).

Processing Visual Logic

To ingest a flowchart or graph, use the process_visual_logic method. This function extracts the logical flow and provides a structured representation that can be injected into an LLM's context window.

Usage Example

from deeptrain import LogicParser

# Initialize the parser
parser = LogicParser()

# Process a flowchart from a local image
graph_data = parser.process_visual_logic(
    source="path/to/workflow_diagram.png",
    output_format="json",
    detail_level="high"
)

print(graph_data)

API Reference: process_visual_logic

| Parameter | Type | Description | | :--- | :--- | :--- | | source | str | Path to the image file, URL, or raw diagram script (e.g., Mermaid). | | output_format | str | The desired structure: "json", "textual_logic", or "adjacency_list". | | detail_level | str | Controls the depth of extraction: "low" (main nodes) to "high" (includes attributes/sub-text). | | ocr_refinement | bool | If True, applies specialized OCR to clarify text within small nodes. |

Returns: A dict containing the logical mapping of the diagram.

Handling Script-Based Diagrams

Deeptrain can interpret diagram-as-code formats directly. This is particularly useful for agents that need to validate architectural decisions or generate documentation from codebases.

mermaid_script = """
graph TD
    A[User Request] --> B{Authorized?}
    B -- Yes --> C[Fetch Data]
    B -- No --> D[Error 403]
"""

# Convert Mermaid script into a logical prompt for an LLM
logic_summary = parser.interpret_script(
    script=mermaid_script,
    provider="mermaid"
)

Logical Output Structure

When processing a visual diagram, Deeptrain returns a structured object. This allows your AI agent to understand not just the labels, but the connectivity of the information.

Sample JSON Output:

{
  "diagram_type": "flowchart",
  "nodes": [
    {"id": "n1", "label": "Start", "type": "entry"},
    {"id": "n2", "label": "Process Payment", "type": "action"},
    {"id": "n3", "label": "Success?", "type": "decision"}
  ],
  "edges": [
    {"from": "n1", "to": "n2", "connection": "direct"},
    {"from": "n2", "to": "n3", "connection": "conditional"}
  ],
  "logic_path": "n1 -> n2 -> n3"
}

Best Practices

• Resolution: For image-based flowcharts, ensure a minimum resolution of 720p for accurate text extraction from nodes.
• Contrast: High-contrast diagrams (dark text on light backgrounds) yield the highest accuracy for relational mapping.
• Prompting: When feeding this data to an LLM, use the textual_logic output format to help the model "visualize" the steps sequentially.