Crate product_farm_llm_evaluator 

LLM-based Rule Evaluation for Product-FARM

This crate provides LLM (Large Language Model) based rule evaluation, allowing rules to use AI reasoning instead of deterministic JSON Logic.

Features

• LlmEvaluatorConfig - Configuration for LLM calls (model, temperature, prompts)
• ClaudeLlmEvaluator - Anthropic Claude implementation (requires anthropic feature)
• OllamaLlmEvaluator - Ollama local LLM implementation (requires ollama feature)
• PromptBuilder - Build context-rich prompts with rule metadata
• ParallelLlmExecutor - Execute LLM rules in parallel
• RuleEngineLlmConfig - Environment-based configuration

Environment Variables

All configuration can be loaded from environment variables with the RULE_ENGINE_ prefix. See env_config module for full list.

Quick Start

# Use Ollama (default)
export RULE_ENGINE_LLM_PROVIDER=ollama
export RULE_ENGINE_OLLAMA_MODEL=qwen2.5:7b

# Or use Anthropic
export RULE_ENGINE_LLM_PROVIDER=anthropic
export RULE_ENGINE_ANTHROPIC_API_KEY=your-key

Example

use product_farm_llm_evaluator::{
    LlmEvaluatorConfig, RuleEngineLlmConfig,
    PromptBuilder, RuleEvaluationContext, AttributeInfo,
    ParallelLlmExecutor, ParallelExecutorConfig,
};

// Load config from environment
let config = RuleEngineLlmConfig::from_env();
println!("{}", config.summary());

// Build a context-rich prompt
let context = RuleEvaluationContext::new("calculate-premium")
    .with_description("Calculate insurance premium")
    .add_input(AttributeInfo::new("age").with_description("Driver's age"))
    .add_output(AttributeInfo::new("premium").with_description("Monthly premium"));

let prompt = PromptBuilder::new().build(&context);

Re-exports

pub use env_config::RuleEngineLlmConfig;

pub use env_config::GlobalLlmConfig;

pub use env_config::AnthropicConfig;

pub use env_config::OllamaConfig;

pub use env_config::RetryConfig as EnvRetryConfig;

pub use env_config::ENV_PREFIX;

Modules

env_config

Environment-based configuration for LLM evaluators.

Structs

AttributeInfo

Metadata about an attribute for prompt context

ClaudeLlmEvaluator

LlmEvaluatorConfig

Configuration for LLM-based rule evaluation.

LlmRuleResult

Result of a single LLM rule evaluation

OllamaLlmEvaluator

ParallelExecutorConfig

Configuration for the parallel executor

ParallelLlmExecutor

Async parallel executor for LLM rules

PromptBuilder

Builds prompts for LLM evaluation with rich context

RetryConfig

Retry configuration with exponential backoff

RuleEvaluationContext

Full context for LLM rule evaluation

RuleMetadata

Metadata for rule evaluation

Enums

LlmEvaluatorError

Errors that can occur during LLM evaluation

OutputFormat

Output format expected from the LLM

OutputFormatInstructions

Functions

default_system_prompt

Generate a system prompt for rule evaluation

Type Aliases

LlmEvaluatorResult