Job descriptions often contain vital data buried in dense paragraphs of text—information like donor requirements, nationality restrictions, or contract types. For my project on Humanitarian Sector Employment Dashboard I created an n8n workflow designed to automatically read, analyze, and structure this data using Large Language Models (LLMs).
What This Workflow Does
The Relief Jobs Analysis workflow acts as an autonomous data analyst. It wakes up on a schedule, identifies job postings that have not yet been analyzed, and sends their descriptions to an AI model.
The AI converts the unstructured markdown text into a clean JSON object containing specific metadata (such as whether a job is “project-based” or restricted to “nationals only”). Finally, the workflow saves this enriched data back into a PostgreSQL database, likely to power a dashboard or search filter.
How It Works: The 3-Step Architecture
The workflow operates in a linear loop consisting of ingestion, analysis, and storage.
1. Smart Ingestion (The Schedule & SQL)
The process begins with a Schedule Trigger that runs every 3 hours. Instead of grabbing every job in the database, the workflow uses a smart SQL query to ensure efficiency.
It connects to a Postgres database and selects only jobs (job_id and body_markdown) that exist in the jobs table but do not yet exist in the job_llm_extractions table. This “diff check” prevents the AI from re-analyzing the same job twice, saving on API costs and processing time.
2. The AI Analyst (LangChain & OpenAI)
The core intelligence lives in the Information Extractor node, powered by an OpenAI Chat Model (specifically gpt-4.1-mini as configured).
The workflow feeds the job description text to the AI with a specific system prompt: “You are a highly intelligent and precise data extraction assistant specializing in analyzing job descriptions from the humanitarian and development sectors”.
The AI is instructed to return a strict JSON schema containing the following insights:
-
Donors: Extracts specific entities like USAID or ECHO.
-
Nationalization: Determines if the role is for local nationals only and provides reasoning.
-
Contract Logic: Identifies if the role is project-based or on a rolling basis.
-
Seniority: Infers the level (e.g., “Managerial”) based on years of experience required.
-
Languages: Splits languages into “Mandatory” vs. “Preferred”.
3. Structured Storage (Postgres Upsert)
Once the AI returns the structured JSON, the Insert or update rows in a table node takes over. It maps the AI’s output fields (like seniority_level, mentioned_donors, and is_project_based) directly to columns in the job_llm_extractions table.
The node uses an “Upsert” operation based on the job_id. This means if a record for that job somehow already exists, it updates it; otherwise, it creates a new entry, ensuring database integrity.
Why This Matters
By standardizing these data points, this workflow allows a humanitarian organization to move from simple keyword searching to complex filtering—enabling queries like “Show me all Managerial WASH positions funded by USAID that are open to international staff.”
Would you like me to…
-
Create a SQL script to generate the
job_llm_extractionstable schema so it matches the fields expected by this workflow? -
Help you adjust the system prompt to extract additional specific fields (like “Remote Policy” or “Salary Range”)?