Observability & AI Safety
This lesson covers monitoring AI systems with Langfuse and protecting against safety risks.
Like a doctor monitoring vital signs — AI observability tracks latency, cost, and errors in the LLM to spot problems before users feel them.
- Langfuse
- An LLM observability platform that records traces, latency, and cost per call.
- Prompt injection
- A user's attempt to override the system prompt through the user message.
- PII leakage
- Leaking personally identifiable information (names, email, credit cards) through AI responses.
- Cost abuse
- Abuse of the API by issuing excessive requests that cause uncontrolled costs.
