CrackML by @ml.with.umang
GenAI interview guides / LLM Observability & Reliability Interview Guide
GenAI / LLM

LLM Observability & Reliability Interview Guide

Monitor generative AI systems across traces, quality, tools, retrieval, latency, cost, safety, and drift.

Trace the system

Link request → prompt/version → retrieval → model → tools → retries → output → user outcome.

Operational metrics

Track availability, queueing, TTFT, inter-token latency, end-to-end latency, tokens, cost, tool errors, retrieval latency, and fallbacks.

Quality telemetry

Sample groundedness, task completion, safety, corrections, escalation, abandonment, and judge/human scores.

Release discipline

Version every moving part, canary changes, define rollback triggers, and preserve production failures as tests.