Continuum

Every major software failure carries a lesson. When companies publish postmortems — detailed accounts of what broke, why, and what was learned — they contribute to a shared body of engineering knowledge that the entire industry can learn from.

The problem is fragmentation. These postmortems live scattered across hundreds of engineering blogs, status pages, and documentation sites. Finding them requires knowing where to look. Comparing incidents across companies is nearly impossible.

Continuum solves this. It is a single, structured repository of industry postmortems — aggregated automatically, searchable semantically, and always linking back to the original source.

The original author's words are never altered. The raw narrative, the technical diagrams, the candid timelines — they remain exactly as published. AI is used only as a navigational aid: summaries and root cause categorization appear in a sidebar, never replacing the source material.

Source Mirroring

The original incident report is always the primary source of truth. Continuum preserves it exactly — including the author's framing, phrasing, and technical context.

Blameless by Design

Following Google's SRE principles and Etsy's postmortem culture, Continuum treats every incident as a systemic learning opportunity, not an individual failure.

AI as Navigator

AI-generated summaries and root cause categories exist to help you find the right postmortem faster. They are explicitly secondary to the original text and clearly labeled as such.

RSS feeds and public status pages are monitored continuously for new incident reports and engineering blog posts from tracked companies.

Modular Python Incident Handlers — one per company — fetch metadata and content. Each handler is purpose-built for its source's structure.

Raw content is committed directly to a GitHub repository (docs-as-code). Metadata is stored in a Postgres database. The Git history is the audit trail.

An LLM generates a concise 120–140 word summary and predicts a root cause category using chain-of-thought prompting grounded in the original text.

FastText embeddings convert each postmortem into a numeric vector, enabling semantic search — find similar incidents across different companies by context, not just keywords.

Conceived, designed, and built Continuum — from the data collection architecture and AI enrichment layer to the brand identity and frontend.

Originated the git-scraping technique — the foundational data collection pattern behind Continuum's handler architecture.

Demonstrated git-scraping at scale across dozens of cloud providers. Their modular, per-company repo pattern directly inspired Continuum's handler design.

The academic research paper on AI-assisted root cause analysis that shaped Continuum's AI enrichment layer — including root cause categorization, chain-of-thought prompting, and FastText embeddings.

The SRE book's chapters on postmortem culture, blameless retrospectives, and action item frameworks defined the principles Continuum is built to serve.

Pioneered open postmortem culture in the tech industry. The Morgue open-source tool and their Debriefing Facilitation Guide provided early architectural reference.

Python

Handler scripts, backend logic

FastAPI

REST API

Next.js 16

Frontend — public site + admin

Supabase (Postgres)

Metadata storage

ChromaDB

Vector search

FastText

Semantic embeddings

Groq / Ollama

LLM summarization

GitHub Actions

Handler automation (cron)

Vercel

Frontend hosting

Inter

UI typeface

JetBrains Mono

Monospace / code font