5 specialisations · 112 tools
Start from what you own.
The stack diagram answers where things sit; this answers what you are accountable for. Each specialisation lists the tools that role actually owns, grouped by the layer they live in — so you can see both at once, and see which cross-layer tools (the gateway, the eval harness, the orchestrator) land in more than one column.
ML Platform
24 tools · 3 layersThe paved road other engineers build on.
Asks How does a model get called, and what does a teammate inherit?
AI Infrastructure
18 tools · 4 layersTokens per second, per dollar, per GPU.
Asks Why is this slow, and what does the next hardware change buy me?
Data & Retrieval
23 tools · 4 layersWhat the model knows, and whether it is the right thing.
Asks Why does it know the wrong thing, or nothing?
Applied Engineering
46 tools · 6 layersThe feature, end to end, in the product.
Asks How do I get from a model call to a working surface?
Production & Governance
25 tools · 5 layersWhether it can be trusted, and whether it can be let go.
Asks How do I know it still works, and what am I accountable for?
Roles are assigned by what the tool is for, not by who vendors it, and a tool can belong to two — an eval framework really is part of what a platform engineer owns and part of what an applied engineer owns. That makes these counts overlap by 24 entries, so the five totals sum to 136 rather than 112; that is the honest reading, not a rounding error. See the method for how the index decides anything at all.