Manual Import Extractor (How It Works)

This is a walkthrough of the manual job import flow, which allows users to add jobs that aren't captured by automated scrapers.

Big Picture

Instead of scraping a website, the manual extractor takes a raw job description (pasted text), parses the details (using AI), and allows the user to review and edit the data before importing it into the pipeline.

1) Input

The user provides input via the Manual Import sheet in the UI. They paste a full job description, copied from any source (job board, company site, email, etc.).

2) AI Inference

When the user clicks "Analyze JD", the orchestrator calls an internal endpoint (/api/manual-jobs/infer).

The server-side service (orchestrator/src/server/services/manualJob.ts) then:

Sends the raw text to an LLM (via the configured provider; OpenRouter by default).
Uses a specific prompt to extract structured data (title, employer, location, salary, etc.).
Returns a JSON object containing the inferred fields.

If the LLM API key is not configured (e.g. LLM_API_KEY for OpenRouter/OpenAI/Gemini), the inference step skips and warns the user to fill details manually.

(OPENROUTER_API_KEY is deprecated and is automatically copied to LLM_API_KEY for compatibility.)

3) Review and Edit

The inferred data is populated into a form in the UI. The user can:

Correct any mistakes made by the AI.
Add missing information.

4) Storage and Scoring

Once the user clicks "Import Job", the data is sent to /api/manual-jobs/import.

The orchestrator:

Generates a unique ID for the job if no URL is provided.
Saves the job to the database with the source set to manual.
Asynchronously triggers scoring: The job is immediately run through the suitability scorer (orchestrator/src/server/services/scorer.ts) against the user's current resume profile.
Updates the job record with the suitability score and reason once complete.

1.9 KiB Raw Blame History