The Knowledge Gardener

For the owner of a large knowledge base that grows every day.

One signal · 10 specialists · one result

Keeps knowledge3
Runs infrastructure3
Audits and guards4

How the work moves

Six steps, each done by one specialist

Starts with

New entries proposed for the knowledge base

First

1Accepts, holds or rejects each entry against the existing topicsKnowledge Entry Screener 2Scores the health of every topicKnowledge Base Health Analyst 3Reports the topics that have driftedKnowledge Drift Detector

Then

4Proposes entries to move from crowded topics as a dry runKnowledge Rebalancer 5Prepares new material so it can be searchedEmbedding Backlog Clerk 6Clears old stored versionsStorage Space Reclaimer

Ends with

A balanced, searchable knowledge base without old clutter

the signalthe resultEach step is done by one specialist on the crew

The story

Each proposed entry is tested against the existing topics and accepted, held or rejected. The agent scores every topic's health and reports the ones that have drifted. Where some topics are overcrowded, it proposes which entries to move and shows the plan as a dry run. New material is embedded so it can be searched, and old stored versions are cleared.

The crew

The specialists behind this role, grouped by the kind of work they do.

Audits and guards

Knowledge Entry ScreenerTests each proposed entry against the existing topics and records an accept, hold or reject decision with the measurement behind it.Also serves The Model Interviewer
Knowledge Base Health AnalystThis agent scores every topic on dozens of health measures and reports which topics need more depth or rebalancing.Also serves The Model Interviewer
Knowledge Drift DetectorRuns a deeper clustering analysis on selected topics and reports those whose content has drifted from where it started.
Screening Rule WriterWrites the rules that tighten or relax acceptance for particular topics and switches individual rules on and off.

Keeps knowledge

Knowledge RebalancerFinds the overcrowded topics, proposes which entries to move at a cautious, balanced or aggressive setting, and shows the plan as a dry run before anything moves.
Vector Library CuratorKeeps a large searchable collection healthy: it embeds new material, rebuilds and checks the search indexes, scores how well each topic cluster hangs together on the GPU, accepts or rejects new clusters, rebalances crowded ones, and retires stale index files.Also serves The Model Interviewer
Knowledge Base MaintainerThis agent runs the regular upkeep jobs: screening new entries, rebalancing between related topics, refreshing embeddings and rebuilding the search index.Also serves The Model Interviewer

Runs infrastructure

Embedding Backlog ClerkFinds the collections with material waiting to be embedded, queues the work and starts the workers.
Storage Space ReclaimerLists the old stored versions, confirms which one is active and clears the ones no longer needed.Also serves The Housekeeper
Embedding Migration OperatorRebuilds the search vectors for a whole index or for all engineering notes, in bulk on a GPU or one item at a time on CPU.Also serves The Search Keeper

Where this crew also works

The Model Interviewer44

Each bar counts the specialists this role shares with another.

The engineering behind this role

GPU vector search and corpus analyticsCapability Redistribution inside the databaseEngineering story From health scores to decisionsEngineering story All capabilitiesThe engineering, by what it does