Setting Up a Local Subagent to Filter Remote Job Feeds
Ads
Checking job boards every morning burns energy before you write an application. Most search alerts return roles with mismatched tech stacks, hidden location restrictions, or experience requirements buried at the bottom of the page.
Setting up a local subagent job filter automates this filtering step. By sending public RSS feeds from remote boards through a structured prompt, you can generate a short list of matched roles every day without paying for job aggregator subscriptions.
The Subagent Architecture Explained
You do not need an orchestration framework like LangChain to build a job scanner. A working script needs three simple parts:
Using a strict schema forces the model to evaluate listings against your requirements as clear pass/fail rules.
The Structured Screening Prompt
Pass each posting through this prompt to get a structured evaluation:
You are a precise data extraction subagent evaluating a job posting for a candidate. The candidate's hard constraints are: 1. Remote location: Must permit US/Canada or Global remote work. Exclude listings requiring on-site presence or hybrid attendance in a specific city. 2. Required core stack: Must include Python OR Go. 3. Experience boundary: Must not require more than 6 years of experience. Evaluate the job listing text and respond ONLY in valid JSON matching this structure: { "is_match": boolean, "rejection_reason": "string or null", "extracted_title": "string", "company": "string", "salary_mentioned": "string or Unlisted", "primary_stack": ["string"] } Do not provide conversational commentary. Output valid JSON only.Running the Triage Procedure Tonight
You can run this locally using Python and an API endpoint or a local runner like Ollama:
This narrows your search to five to eight qualified roles a day rather than dozens of mismatched posts.
Frequently Asked Questions
Can this script run with a free local model?
Yes. Small open-weights models running in Ollama handle JSON extraction well. Because you are classifying data rather than writing prose, an 8-billion parameter model is enough.
Will parsing job feeds get my IP address blocked?
Public RSS feeds exist for automated consumption. Keep requests reasonable: fetch updates once or twice daily, supply a clean User-Agent header, and do not scrape pages behind login screens.
Why use JSON schema enforcement instead of standard text?
Unstructured text outputs vary from run to run, breaking automated workflows. Strict JSON ensures your script can write the matching jobs directly into your tracking file without manual fixes.
