AI Role Taxonomy Engine.

The AI Role Taxonomy Engine is Orbyt's canonical catalog of the specialized roles the U.S. labor market keeps renaming. As of August 2026 it tracks 598 distinct roles, from role:ai-engineer and role:ml-engineer to emerging roles like role:llm-systems-engineer, role:ai-agent-engineer, and role:rag-platform-engineer that did not exist three years ago.

It exists because the labor market does not agree on names. The same job is called Machine Learning Engineer, ML Platform Engineer, AI Engineer, AI Infrastructure Engineer, Senior ML Engineer, and AI Systems Engineer at six different companies on the same week. Without a canonical layer, every downstream query: salaries, skill premiums, hiring velocity, projections: would silently pivot on the noun the customer happened to type.

Every other Orbyt engine resolves through this one first. When you ask analyze_compensation for an "AI engineer in San Francisco," the Role Taxonomy Engine runs first. It returns role:ai-engineer with a confidence score, a list of alternates that nearly matched (role:ml-engineer, 0.82 similarity), and a methodology version. Every subsequent call uses the stable slug so the answer is reproducible.

That is the difference between a tool that answers your question and a tool you can build a business on. Data you can audit starts with stable identifiers. So does ours.

The catalog is organized into eight role families. Each family clusters related roles by core competency, then by specialization within that family.

Adding a new role requires three independent signals: at least 200 distinct job postings using the title across the past 90 days, at least one comp-survey reference (Levels.fyi, Pave, or a published industry survey), and an H-1B LCA filing (DOL) using the title. This three-source threshold keeps the catalog from inflating with marketing vocabulary while still letting genuinely new categories enter quickly. The current rate of new role additions runs at 6-8 per quarter.

Deprecating a role takes the opposite path. We never delete a role from the live API. Backwards compatibility is sacred. Instead, deprecated roles carry a successor_slug field, and lookups against the deprecated slug return a 200 with a deprecation_notice in the envelope. The role continues to resolve forever.

Each role lives at the intersection of three orthogonal hierarchies: family, seniority, and specialization. The taxonomy stores them as separate fields rather than flattening into one string, because each one moves independently.

Three hierarchies, three queries. By family: list every role in thellm-and-agents family to map a team. By seniority: get the salary band forstaff across every role family to benchmark a leveling conversation. By specialization: compare platform roles across families to understand the cross-domain premium for platform work. The taxonomy supports all three without any role-string magic.

Seniority multipliers are a locked calibration curve applied to the role's modelled low, median and high. The engine does no per-seniority percentile tagging, which its own source states, so no level maps to a published percentile. The multiplier is the engine's output; the customer multiplies the role's national median to produce the band figure.

The Bureau of Labor Statistics maintains the Standard Occupational Classification system covers 800-ish occupation categories covering the entire U.S. labor market. SOC is the authoritative federal taxonomy. Every Orbyt role maps to one or more SOC codes so customers can cite a government data source for every salary figure we publish.

The mapping is two-tier: a primary SOC code that drives the salary baseline, and an optional secondary code that provides additional signal for ambiguous roles. For role:ai-engineer, the primary is 15-2051 (Data Scientists and Mathematical Science Occupations) because BLS does not yet have a dedicated SOC code for AI Engineers. The secondary is 15-1252 (Software Developers) because the role overlaps both categories.

Where BLS lacks a SOC code (most modern AI roles), we apply a documented premium adjustment on top of the primary SOC baseline. The adjustment is derived from the role's H-1B LCA distribution relative to the SOC's overall distribution. For AI Engineer in 2026, the H-1B LCA median for the role-specific filings runs ~22% above the SOC 15-2051 median: that 22% becomes the premium multiplier.

Every Orbyt response that returns salary data includes the SOC code, the SOC name, the premium multiplier, and the methodology version. The /lineage/[data_point_id] endpoint returns the full provenance trail: every source that fed into this specific salary, with timestamps and reconciliation rules. That is end-to-end traceability applied to U.S. labor data.

The Role Taxonomy Engine ships on the free tier. It is the substrate every other engine builds on, so we keep it at the floor price: a starting line that pays for its own compute without locking the ecosystem out. The free tier caps reads at 60 per minute per API key, 1,000 requests a month.

Restricted API keys with per-engine scopes are available on all paid tiers. Grant a key only the scopes it needs (skills:read for a learning tool that does not need company data, company_data:read for a comp-consulting tool that does not need skill premiums, etc.). The full pricing page lives at /orbyt-intelligence/pricing.