Series

Building Nightwatch: A Production-Ready Multi-Agent AI On-Call System

A hands-on engineering series on building Nightwatch, a production-ready multi-agent AI on-call partner using LangGraph, Python, CloudWatch, PagerDuty, FastAPI, and Next.js.

We’ll design the system from first principles, starting with the architecture and agent state, then build the Watcher, Analyzer, Fixer, human-in-the-loop approval flow, Postmortem agent, orchestration layer, PagerDuty integration, real-time dashboard, and production hardening.

This series is for backend engineers, SREs, DevOps engineers, and AI engineers who want to move beyond toy agents and understand how to build agentic systems that are observable, typed, interruptible, cost-aware, and safe enough to run near production.

Stop Paging Humans: Designing an AI On-Call Partner with LangGraph
Part 1 of Building Nightwatch, a production-ready, multi-agent AI system that handles incidents while you sleep.
May 17, 202610 min read
Building the Eyes of Your Infrastructure: AWS CloudWatch Anomaly Detection in Python
Part 2 of Building Nightwatch, turning logs and metrics into actionable incidents, without breaking the bank.
May 20, 202610 min read2
Watching with Intent: The Rules System and Custom Signal Sources
Supplementary, Configuring what Nightwatch watches without writing Python. YAML-driven incident detection.
May 20, 20269 min read7
The Analyzer Agent: Turning Logs into Root-Cause Hypotheses with LLMs
Part 3 of Building Nightwatch — structured outputs, token budgets, and calibrated confidence.
Jul 1, 20265 min read
The Fixer Agent: Human-in-the-Loop Remediation with LangGraph's interrupt()
Part 4 of Building Nightwatch — durable pauses, severity-based auto-approval, and the executor pattern.
Jul 2, 20265 min read

Command Palette