LLM safety, guardrails, prompt injection, PII redaction, Hinglish NLP, Indian languages, small language models