FlexiRule Docs

AI Features

Leveraging Artificial Intelligence within FlexiRule logic.

Improve this page

AI Features#

Keywords: AI, machine learning, prompt engineering, automation, intelligent rules

Audience#

End Users
Developers

Overview#

FlexiRule integrates AI capabilities directly into the visual orchestration layer, allowing for “intelligent” business processes that can analyze text, make predictions, or generate content as part of a rule execution.

When to Use#

Use AI Features when you need to handle unstructured data (e.g., sentiment analysis of comments).
Use AI for complex decision-making that is difficult to express in traditional boolean logic.
Use AI for automated content generation based on document state.

Do Not Use#

Do not use AI for deterministic logic where a simple condition or math formula suffices.
Do not use AI for high-volume, low-latency requirements without considering cost and performance.

Visual Example#

graph TD A[Start] --> B{AI Classifier} B -- "High Priority" --> C[Alert Manager] B -- "Low Priority" --> D[Log Activity] C --> E[End] D --> E

Capabilities#

1. Context-Aware Prompting#

AI Actions can ingest the entire execution context (doc, vars) to generate responses tailored to the specific record state.

2. Natural Language Classification#

Classify documents or inputs into predefined categories without writing complex regex or lookup tables.

3. Data Extraction#

Extract structured information from unstructured text fields and map them to document fields or variables.

AI Integration Flow#

sequenceDiagram participant E as Engine participant A as AI Action participant P as AI Provider (LLM) E->>A: Execute Node (Context) A->>A: Build Prompt with doc/vars A->>P: Request (Prompt) P-->>A: Response (JSON/Text) A-->>E: Update Context (vars.ai_result)

Prompt Best Practices#

Be Specific: Provide clear instructions and examples (few-shot prompting) within the action configuration.
Limit Output: Request the AI to return data in a specific format (e.g., JSON) for easier parsing by subsequent nodes.
Security: Be mindful of PII (Personally Identifiable Information) when sending data to external AI providers.

Limitations & Considerations#

Latency: AI requests are network-bound and can introduce delays in rule execution.
Cost: Each AI execution incurs a cost based on the provider’s token usage.
Hallucinations: AI outputs should be validated, especially when performing critical mutations.

Was this page helpful?