Skip to content
Docs

AI Features Overview

WP Career Board can use AI across the hiring flow: job description writing, natural-language job search, candidate-to-job matching, applicant ranking with summaries, and cover-letter drafting. All AI features ship in the Pro plugin. The Free plugin defines the gate filters (wcb_ai_description_enabled, wcb_ai_ranking_available, wcb_ai_matching_available) and the UI surfaces, so Pro wires the AI in without changes elsewhere.

This page summarises what’s in the current release (1.4.3), what’s gated behind an AI provider being configured, and how the pieces fit together. If you’re on Free, treat this as a feature preview: the flows here are what you get when you upgrade.

What ships today (1.4.3)

Section titled “What ships today (1.4.3)”

The AI feature set matured in Pro 1.3.0 and has shipped in every release since. All of the following are live in code:

Feature Surface Who uses it
Job embeddings Background: every new job is embedded on wcb_job_created into the wcb_ai_vectors table. System
Index existing jobs (backfill) “Index existing jobs” button under Settings -> AI Settings embeds jobs that existed before AI was enabled. Site owners
AI Chat Search block wcb/ai-chat-search block (a chat-style search box). An AI Job Search page carrying this block is auto-created when Pro sets up. Closest-matched jobs come back via POST /wcb/v1/ai/match. Candidates
AI candidate matches (“Recommended for you”) Top-N jobs matched to a candidate’s resume, available via GET /wcb/v1/candidates/{id}/matches and surfaced for the current candidate. Candidates
AI Job Description Writer “Generate with AI” button on the post-a-job form, gated behind wcb_ai_description_enabled. Sends title + company type + location to the provider; returns structured HTML. Employers
AI applicant ranking One-click “Rank by AI fit” on the Employer Dashboard. Each applicant gets a 0-100 fit score, a one-line reason, and a TL;DR summary, sorted best-first via GET /wcb/v1/ai/ranked-applications/{job_id}. Employers
Applicant TL;DR summaries A one-to-two-sentence neutral summary of each applicant’s background, shown alongside the fit score in the dashboard list and detail. Employers
Auto-score on submit Optional. When enabled, new applications are scored in the background (cron) so the dashboard shows AI fit instantly. Employers
AI cover-letter writer In the job apply panel, candidates generate a tailored cover letter from their resume and the job, edit it, then submit via POST /wcb/v1/jobs/{job_id}/ai-cover-letter. Candidates

In Free, every path is gated. The Chat Search block renders nothing for visitors (and a configure hint for admins), the description and cover-letter buttons are hidden, the ranking controls don’t appear, and the AI REST routes aren’t registered. No broken UI surfaces in Free.

The two-provider model

Section titled “The two-provider model”

AI is configured per task, not as a single global provider:

  • Analysis & ranking (completions) drives the description writer, applicant ranking, TL;DR summaries, the cover-letter writer, and the chat search assistant. Choose Anthropic Claude, OpenAI, or Ollama.
  • Embedding & matching drives job embeddings, AI Chat Search, and candidate matches. Choose OpenAI or Ollama (Claude has no embeddings API, so it is not an embedding option).

Each provider has its own API key (or base URL for Ollama) and its own model selection. You can run, for example, Claude for ranking copy and OpenAI for embeddings at the same time. See 02-setup-and-providers.md.

What gets cached and re-billed

Section titled “What gets cached and re-billed”

Applicant fit scores, reasons, and summaries are cached in application post meta (_wcbp_ai_fit_score, _wcbp_ai_fit_reason, _wcbp_ai_summary, _wcbp_ai_scored_at). Re-opening the dashboard never re-bills the model; ranking computes only what is missing. Force a re-score by passing $force to AiModule::score_application() from a custom integration.

Embeddings are stored once per job in wcb_ai_vectors and reused for every search. Query embeddings (chat search, candidate match) are computed per request and not stored.

Free vs Pro at a glance

Section titled “Free vs Pro at a glance”
Free Pro
Job posting form Full editor Full editor + AI description writer button
Candidate dashboard / resume Resume Builder, manual profile fill Same + resume feeds AI matching, ranking, and cover letters
Search bar Keyword search, taxonomy filters Keyword + filters + AI Chat Search block + AI Job Search page
Apply panel Cover-letter field (manual) Same + “Generate with AI” cover-letter button
Applications screen / dashboard List, status filter, manual review Same + AI fit score, reason, TL;DR, “Rank by AI fit”, optional auto-score
AI Settings tab Hidden Visible under WP Admin -> Career Board -> Settings -> AI Settings
Database Standard tables only Adds wcb_ai_vectors for embeddings

When AI is worth turning on

Section titled “When AI is worth turning on”

Turn it on if:

  • Your board has more than ~30 active jobs and candidates struggle to find the right one with keyword search alone.
  • You want the description-writer assist to speed up job posting.
  • You triage many applicants per role and want a fit-score plus TL;DR to prioritise reviews.
  • You want candidates to draft cover letters from their resume in one tap.

Skip it if:

  • You’re under 20 active jobs and a handful of applications a week. The description writer might still help, but matching and ranking won’t move the needle yet.
  • You can’t share job / candidate text with a third-party LLM provider for privacy / compliance reasons. (Use Ollama locally; see 02-setup-and-providers.md.)

How it works under the hood

Section titled “How it works under the hood”

When AI is enabled and the relevant provider is configured:

  1. On job publish: Pro hooks wcb_job_created and calls AiModule::generate_job_embedding(). The job title + content are embedded by the configured embedding provider and stored as a binary-packed float vector in wcb_ai_vectors.
  2. On AI Chat Search: the candidate’s query is embedded the same way, then cosine-similarity-compared against every stored job vector. Top 10 matches are returned via POST /wcb/v1/ai/match.
  3. On candidate matching: the candidate’s resume text (built by the Pro Resume module via the wcbp_candidate_resume_data filter) is embedded and matched against job vectors via GET /wcb/v1/candidates/{id}/matches.
  4. On description writer: the form sends title + company type + location to POST /wcb/v1/jobs/ai-description. The completion provider returns a structured-HTML description; the editor inserts it.
  5. On applicant ranking: each application’s job title + candidate resume text is sent to the completion provider with a JSON scoring prompt. Returns {score, reason, summary} per application, cached in post meta. GET /wcb/v1/ai/ranked-applications/{job_id} returns the list sorted best-first.
  6. On auto-score (optional): when wcbp_ai_auto_rank is on, Pro hooks wcb_application_submitted and schedules a background wcbp_ai_score_application cron event so the score is ready before the employer looks.
  7. On cover-letter generation: the candidate’s resume text + the job are sent to the completion provider via POST /wcb/v1/jobs/{job_id}/ai-cover-letter; the returned draft is inserted into the cover-letter field for editing.

Embedding paths go through the configured embedding driver; completion paths through the configured completion driver. Both drivers are OpenAI / Anthropic Claude / Ollama (or one registered via wcbp_ai_provider_drivers).

What gets sent to the AI provider

Section titled “What gets sent to the AI provider”

The privacy-relevant breakdown:

Feature What’s sent
Job embedding (on publish / backfill) The job title and description as plain text.
AI Chat Search The candidate’s typed query string.
AI candidate matches The candidate’s flattened resume text.
AI Description Writer The job title, company type, and location.
Applicant ranking / TL;DR The job title and the applicant’s flattened resume text.
AI cover letter The job title, a job-content excerpt, and the candidate’s resume text.

For privacy-sensitive deployments, run Ollama locally - see 02-setup-and-providers.md. Nothing leaves your server.

What it doesn’t do

Section titled “What it doesn’t do”
  • No auto-rejection. Ranking returns scores; nothing acts on them automatically. Every decision still lands with the employer.
  • No auto-matching emails. Job Alerts use keyword / filter match, not AI semantic match.
  • No content generation outside the writer and cover-letter tools. AI does not auto-fill candidate bios, company “about us” sections, or similar.
  • No data sold or shared. All API calls are between your site and the configured provider directly - Wbcom never sees or proxies the data.

Where to go next

Section titled “Where to go next”