diff --git a/documentation/extractors/README.md b/documentation/extractors/README.md
index f65c101..f6f5196 100644
--- a/documentation/extractors/README.md
+++ b/documentation/extractors/README.md
@@ -4,3 +4,4 @@ Technical breakdowns of how each extractor works.
 
 - Gradcracker: `gradcracker.md`
 - JobSpy: `jobspy.md`
+- UKVisaJobs: `ukvisajobs.md`
diff --git a/documentation/extractors/ukvisajobs.md b/documentation/extractors/ukvisajobs.md
new file mode 100644
index 0000000..eff5986
--- /dev/null
+++ b/documentation/extractors/ukvisajobs.md
@@ -0,0 +1,87 @@
+# UKVisaJobs Extractor (How It Works)
+
+This is a plain-English walkthrough of the UK Visa Jobs extractor. It's the most complex one because the site requires an authenticated session before the API will return jobs.
+
+## Big picture
+
+There are two layers:
+
+1) `extractors/ukvisajobs/src/main.ts` handles logging in, talking to the UKVisaJobs API, and writing a Crawlee-style dataset.
+2) `orchestrator/src/server/services/ukvisajobs.ts` runs that extractor, reads the dataset, de-dupes results, and optionally enriches descriptions.
+
+## 1) Authentication and session cache
+
+The API requires a token + cookies. The extractor keeps these in a cache file:
+
+- `extractors/ukvisajobs/storage/ukvisajobs-auth.json`
+
+Flow:
+
+- If there's a cached session, it uses it.
+- If not, it launches a real browser (Playwright + Camoufox), logs in with `UKVISAJOBS_EMAIL` and `UKVISAJOBS_PASSWORD`, then captures the auth cookies + token.
+- It stores those values in the cache file for reuse.
+
+You can force a refresh with:
+
+- `UKVISAJOBS_REFRESH_ONLY=1`
+
+If the API responds with an "expired" token error, it will automatically re-login and retry.
+
+## 2) API requests
+
+Once authenticated, it posts to:
+
+- `https://my.ukvisajobs.com/ukvisa-api/api/fetch-jobs-data`
+
+Each request:
+
+- Includes the auth token in a form field.
+- Includes cookies in the header (`csrf_token`, `ci_session`, `authToken`).
+- Filters by search keyword if provided.
+- Uses pagination (15 jobs per page).
+
+## 3) Job mapping
+
+The extractor normalizes the raw API data into the project's job shape:
+
+- Salary is built from min/max values and interval.
+- Visa-related flags are turned into a short fallback description if the job has no real description.
+- The `job_link` becomes both `jobUrl` and `applicationLink`.
+
+## 4) Output dataset
+
+The extractor writes the results to:
+
+- `extractors/ukvisajobs/storage/datasets/default/`
+
+It mirrors Crawlee's dataset format:
+
+- One JSON file per job.
+- A combined `jobs.json` containing all jobs.
+
+## 5) Orchestrator flow (how the app uses it)
+
+When the pipeline runs:
+
+- The server spawns the extractor as a child process (`npx tsx src/main.ts`).
+- It can run multiple search terms sequentially (with a short delay between them).
+- It reads the dataset and de-dupes by `sourceJobId` (or `jobUrl` fallback).
+- If a job's description is missing or too short, it makes a direct HTTP request to the job URL and extracts plain text.
+  - This is effectively a curl-style fetch of the job page to fill in the JD for scoring and summarization.
+
+## Controls and limits
+
+Key environment variables:
+
+- `UKVISAJOBS_EMAIL`, `UKVISAJOBS_PASSWORD` (required for auth refresh)
+- `UKVISAJOBS_HEADLESS` (set `false` to show the browser)
+- `UKVISAJOBS_MAX_JOBS` (default 50, max 200)
+- `UKVISAJOBS_SEARCH_KEYWORD` (single keyword filter)
+
+The UI also lets you set max jobs and search terms via the pipeline settings.
+
+## Practical notes
+
+- If you remove the auth cache file, the next run will re-login.
+- The extractor is intentionally polite: it runs low concurrency and adds short delays.
+- If the API or session changes on the UKVisaJobs side, the refresh logic is the first thing to check.