Every AI role gets tagged from a 35-term controlled vocabulary spanning agent / eval / training / inference / modality clusters. Tags that frequently appear together in the same JD pull each other close; thicker edges = more co-occurrences. The clusters that emerge organically are the real sub-disciplines of applied AI engineering right now.
Showing 96 tagged AI roles in Insurance. Layout is a Fruchterman-Reingold force simulation, run server-side to convergence.
Which technologies are hot, which are cooling. Sparkline = 12 weeks of unique roles tagged with each term, last bar on the right is this week. Sorted by absolute pickup. Tags with under 10 lifetime mentions are hidden as noise.
| Tag | 12-week trend | Last 4w | Prior 4w | Δ |
|---|---|---|---|---|
| guardrails | 8 | 6 | ↑+2 | |
| evals | 6 | 4 | ↑+2 | |
| llm_observability | 18 | 17 | ↑+1 | |
| tool_use | 4 | 4 | ·0 | |
| agent_orchestration |
| 17 |
| 19 |
| ↓-2 |
| inference_infra | 5 | 9 | ↓-4 |
| fine_tuning | 4 | 8 | ↓-4 |
| model_serving | 10 | 15 | ↓-5 |
| vector_db | 0 | 5 | ↓-5 |
| rag | 9 | 15 | ↓-6 |