AI Operations Analyst

Production-style agentic incident response for cloud and ML systems.

AI Operations Analyst is a portfolio project focused on operational decision-making rather than chat interaction. It simulates how an on-call analyst monitors production signals, investigates incidents, prioritizes remediation options, and applies only low-risk actions behind explicit approval gates.

What this project demonstrates

• End-to-end incident handling across telemetry review, investigation, recommendation, and controlled remediation.
• Practical agent design with tool orchestration, structured outputs, confidence scoring, and escalation logic.
• Safety-minded automation through approval gates, allowlisted actions, and human-readable reporting.
• Portfolio-ready artifacts that make architecture, scenarios, and engineering tradeoffs easy to review.

What the system does

• Watches logs, metrics, deploys, queue health, and data-quality signals.
• Runs an investigation plan with deploy checks, baseline comparisons, and infrastructure checks.
• Produces a root-cause report with confidence scoring and evidence.
• Suggests ranked actions by cost and blast radius.
• Executes only allowlisted low-risk actions after approval.
• Publishes ticket, PR, and Slack-style summaries automatically.

Stack

• Agent runtime: LangGraph/OpenAI Agents SDK shaped abstractions.
• LLM strategy: GPT-4.1/5-class primary path with cheap fallback hooks.
• Tools: Python functions, SQL, GitHub, Slack, and cloud telemetry connectors.
• Memory: short-term incident state plus runbook lookup hooks.
• Observability: OpenTelemetry-shaped events and evaluation outputs.
• Deployment: Docker-first packaging with AWS-friendly integration points.

Why it is a strong portfolio piece

• Centers on a realistic business problem: reducing incident response time while keeping automation safe.
• Shows engineering restraint by limiting autonomous actions to low-blast-radius changes.
• Demonstrates product thinking through operator-facing summaries, approval workflows, and postmortem-style artifacts.
• Reflects current AI systems work: orchestration, evaluation, safety policy, and integration with operational tooling.

Run it

python -m pip install -e .
ai-ops-analyst --scenario latency_spike
ai-ops-analyst --scenario model_drift --approve-safe-actions

Portfolio artifacts

• Landing page
• Architecture
• Safety policy
• Evaluation dashboard
• Incident scenarios and postmortems

Project layout

• src/ai_ops_analyst/engine.py contains the analyst loop and decision policy.
• src/ai_ops_analyst/runtime.py provides the agent-runtime abstraction.
• src/ai_ops_analyst/connectors.py contains cloud, SQL, GitHub, and Slack connector stubs.
• src/ai_ops_analyst/scenarios.py contains three incident replays.