AI 2 min read STABLE

AI-Ops Assistant in Production: Architecture & Guardrails

Tiered routing, internal knowledge grounding, safe escalation, and when an LLM beats a runbook (patterns from the DailyOps assistant).

D
DailyOps
Published 2026-07-10
A

Context

LLM assistants in operations fail when treated as magic search boxes. Production AI-Ops needs tiered routing, grounded answers, and a clear path to humans. This article documents patterns used by the DailyOps site assistant — applicable to internal NOC copilots and runbook helpers.


When to Use an LLM (and When Not To)

Use LLMPrefer deterministic tooling
Explaining a concept across sourcesExecuting a known remediation script
Summarizing a long runbook for triageParsing structured monitoring alerts
Drafting incident commsEnforcing config compliance (Ansible, OPA)
Routing vague operator questionsCVE version lookups with exact vendor feeds

Rule: if the wrong answer causes outage or compliance breach, do not rely on the LLM alone.


Tiered Routing Model

Route queries by confidence and source type before calling an expensive model.

text
Tier 1 — Site knowledge    → articles, categories, platform FAQ
Tier 2 — Vendor docs       → curated official documentation
Tier 3 — Reasoning LLM     → synthesis when tiers 1–2 insufficient
Tier 4 — Live web search   → CVEs, versions, breaking news (optional)

Benefits:

  • Lower cost and latency for common questions
  • Answers cite sources the operator can verify
  • LLM handles ambiguity only when cheaper paths fail

Grounding on Internal Knowledge

Never let the model invent infrastructure state.

  1. Retrieve — search articles, runbooks, CMDB snippets (RAG)
  2. Compose — LLM answers only from retrieved chunks
  3. Cite — link every claim to a source slug or doc ID
  4. Refuse — if retrieval is empty, say so and offer escalation
text
User question → embed/search internal index → top-k chunks → LLM prompt with citations

Anti-patterns:

  • Dumping entire wiki into the system prompt (stale, expensive, leaky)
  • Letting the model guess hostnames, IPs, or credentials

Guardrails

RiskMitigation
Secret leakageStrip env vars from prompts; block paste of API keys
Harmful commandsRead-only mode; suggest commands, don't auto-run
Hallucinated CVEsTier 4 search + link to NVD/vendor advisory
Off-topic abuseScope system prompt to ops/infra domains
Over-trustShow confidence via sources count, not fake percentages

Escalation path (required):

  • Expert form — email transcript to on-call / platform team
  • Contact fallback — visible when API or model unavailable
  • Rate limiting per IP/session on public widgets

Prompt Design for Ops Personas

Effective ops assistants sound like senior engineers, not chatbots:

  • Short paragraphs, bullet steps for procedures
  • Ask one clarifying question when context is ambiguous
  • Prefer links to DailyOps guides over generic internet advice
  • Admit uncertainty: « I don't have a verified source for that »

Pair with Automation: LLM suggests; Ansible/Terraform executes after human approval.


Observability of the Assistant

Monitor the assistant like any tier-1 service:

MetricWhy
Request rate & latencyCapacity, model slowness
Tier distributionIs routing working?
Escalation rateKnowledge gaps
Error rate (429, 5xx)Quota, upstream failures
Zero-source answersRAG index gaps

Log prompts/responses with PII redaction — retention aligned with security policy.


Validation Checklist

  • Tier 1 answers return without LLM for FAQ-style queries
  • Every LLM answer includes ≥1 verifiable source link
  • Escalation email delivers transcript + visitor contact
  • Fallback message when OPENAI_API_KEY missing or quota exceeded
  • Security review: no secrets in client-side bundles

Key Takeaways

  • AI complements Automation and Observability — it does not replace them
  • Tiered routing keeps cost, latency, and hallucination risk under control
  • Production assistants need escalation, citations, and refusal — not just fluency
Back to articles