I Built a CI/CD Platform From Scratch — From PipelineOS to OdinRun

Description and use

OdinRun is a self‑hosted CI/CD platform evolved from a monolithic runner. It offers reliable execution, real‑time observability, deterministic auto‑remediation, and AI‑assisted failure diagnosis.

Problem

Building a reliable, observable, self‑healing CI/CD runtime from scratch.

Outcome

OdinRun delivers a self‑hosted, observable, and self‑healing CI/CD platform that reduces manual retries, provides real‑time insights, and automatically diagnoses failures.

Workflow

• Decompose monolith into single‑responsibility components
• Fix core runtime bugs (stream demultiplexing, timeouts, cleanup, resource limits)
• Replace synchronous HTTP calls with in‑process domain events
• Add persistence abstractions (SQLite/MongoDB, log/artifact storage)
• Capture real‑time telemetry from Docker daemon
• Expose telemetry via SSE/WebSockets
• Implement deterministic rule‑based auto‑remediation
• Add AI failure diagnosis using structured context

Evidence

The original runner was an ~810-line monolith…subtle bugs plagued the runtime

Source

PIP-33 became the first major reliability milestone. I decomposed the monolithic runner into clean, single-responsibility components

Source

PIP-36: Rule-Based Auto-Remediation…deterministic Rule Engine

Source