LLM Guardrails Implementation

Implement comprehensive guardrails for LLM outputs.

## Application Type
{{application_type}}

## Content Policies
{{content_policies}}

## Compliance Requirements
{{compliance_requirements}}

Build a guardrails system:

```python
class LLMGuardrails:
    async def check_input(self, user_input: str) -> GuardrailResult:
        """Pre-generation checks"""
        pass
    
    async def check_output(self, response: str, context: dict) -> GuardrailResult:
        """Post-generation checks"""
        pass
    
    async def filter_output(self, response: str) -> str:
        """Apply content filtering"""
        pass
```

Guardrail categories:
- PII detection and redaction
- Toxicity filtering
- Topic restriction
- Factuality boundaries
- Format compliance

Include:
- Configurable rule engine
- Async processing for low latency
- Logging and audit trails
- Bypass mechanisms for admin