SkillsRight Intelligence Engine — Overview¶

The SkillsRight Intelligence Engine is an AI platform for executive recruiting. It reads resumes and job descriptions, extracts structured skill profiles, scores candidates against jobs, generates ideal-candidate personas, and produces resume-screening rubrics.

This is an internal staffing tool, not a SaaS product sold to third parties — the company using it is both the technology operator and the employment agency. That posture shapes much of how the system is designed.

What it does¶

Four AI agents, each focused on one domain:

Agent	What it does
Company Profile	Researches a target company, builds a culture archetype, identifies the values and behavioral traits associated with success there.
Job Intelligence	Scores a job description for skills-first hiring quality, extracts required skills, generates an ideal-candidate persona, builds a resume-screening rubric.
Resume Intelligence	Parses a resume (PDF/DOCX/TXT), extracts a structured skill profile (skills, experience, education, certifications), assesses resume quality, infers culture signals from work patterns.
Talent Matching	Scores a candidate against a job — technical fit (skills, experience, role readiness) plus contextual fit (culture alignment, growth trajectory). Includes a deterministic skill-overlap signal grounded in the Lightcast taxonomy.

Who uses it¶

Audience	What they need
Recruiters	Day-to-day evaluation of candidates and jobs. They don't talk to the API directly — they use a frontend app that the engine powers.
App developers	Build the recruiter-facing frontend and any integrations. The engine is API-first — the frontend is a separate codebase that calls the engine's REST endpoints.
Operations / Admin	Run the platform — monitor LLM costs, manage prompts, run the Lightcast catalog sync, watch for failures. Uses the built-in Developer Console at `/dashboard` and `/admin`.
Compliance / Legal	Verify the platform meets EEOC, NYC LL144, CCPA, Colorado AI Act, and related requirements. See the Compliance & Privacy section.

Architectural shape (one paragraph)¶

The engine is a stateless compute service. Resume text or a job description goes in, structured JSON comes out. The consuming frontend application owns all candidate and job data long-term; the engine only keeps operational records (prompt configurations, LLM call audit logs, sync runs). Every AI step is a pure function — JSON in, JSON out — designed to be orchestrated by AWS Step Functions at scale. The engine is provider-agnostic for LLMs: OpenAI, AWS Bedrock, and Anthropic all work, and the choice is configurable per prompt.

For more detail, see Architecture.

Where the data comes from¶

Resumes come from the candidate (uploaded through the frontend app).
Job descriptions come from the staffing company's recruiters (or from job-board ingestion in the frontend).
Skills taxonomy comes from Lightcast — a third-party catalog of 34,000+ canonical skills, refreshed bi-weekly. See Third-Party Integrations.
LLM inference runs against OpenAI / AWS Bedrock / Anthropic, configurable per prompt. See Third-Party Integrations.

How candidates are protected¶

The platform is built with US employment-law compliance as a first-class concern:

The matching prompts never see candidate name, email, phone, location, school name, or employer name — all stripped before scoring.
The persona prompt no longer generates a fictional name, age, or salary — those acted as proxies for protected classes.
The matching prompt no longer scores a retention_potential dimension — tenure patterns correlate with sex/disability/veteran status.
Every LLM call is logged with a fingerprint (not the raw body) by default.
Every score is reproducible from a stored prompt_version + inputs_fingerprint.
Skills extraction can be backed by Lightcast — a deterministic ML extractor — so scoring isn't dependent on the LLM's inference variability.

Full detail in Compliance & Privacy.