Knowledge OS for ops teams
An internal-tools company building a single surface for runbooks, postmortems and SOPs that ops teams actually use. The previous setup: Confluence for docs, Notion for postmortems, Slack for incident notes, a private GitHub wiki for architecture. Knowledge sat in four places and answers lived in zero.
I joined as the first engineer and shipped the v1 in fourteen weeks: a Next.js App Router app with a custom block-based editor, hybrid keyword + vector search, and an embedded answer engine that cites the source paragraph instead of hallucinating one.
Now used by 12 teams inside the design partner. Mean time to find a runbook went from 4 minutes to 18 seconds. The product is live with three more design partners onboarding.
Tech stack
The whole picture.
Top-level technologies. Specific packages and integrations are listed below.
Architecture
End to end.
- Next.js App Router with Cache Components for doc reads
- Streaming answer UI via Vercel AI SDK + Suspense
- Realtime collaboration via Yjs over WebSocket
- Postgres with pgvector for hybrid search
- Embedding pipeline as BullMQ workers (insert/update/delete)
- R2 for asset storage with signed URLs
Libraries
Packages worth naming.
- Tiptapblock editor with custom nodes
- Drizzle ORMPostgres + pgvector access
- tRPCend-to-end typed API
- Vercel AI SDKstreaming answer composition
- BullMQembedding and re-index jobs
Integrations
External systems.
- OpenAIembeddings + answer generation
- Anthropiclong-context summarization
- Slack/ask command + notifications
- Linearauto-link runbooks to incidents
- GitHubauto-pull architecture docs
Building something similar?
Tell me about the product. I'll come back with thoughts and a first shipping date.