Skip to content

WALKTHROUGH.md

Latest commit

 

History

History
229 lines (158 loc) · 6.86 KB

WALKTHROUGH.md

File metadata and controls

229 lines (158 loc) · 6.86 KB

JOB — Walkthrough

A step-by-step guide to running your first complete cycle: scan → draft → report.

Before you begin

You need:

  • Node.js 18+
  • NETROWS_API_KEY — get one at netrows.io (€49/mo, 10k credits)
  • AIAS_API_KEY — your AiAssist Secure key (for resume extraction + company summarisation + email drafts)
  • A copy of your resume as resume/resume.docx
cd job
npm install

Step 1 — Configure your targets

Open targets.yaml. Two sections:

profile — who you are and what you're looking for:

profile:
  name: Mark Allen Evans
  title: Developer Relations / AI Platform Engineer
  skills: [AI, MCP, APIs, DevRel, Node.js, Python, platform engineering]
  target_roles: [developer advocate, developer relations, devrel, developer experience]
  remote_ok: true
  deal_breakers: [no sponsorship, on-site only, intern, internship]

deal_breakers are checked against the full job title + description. Any match drops the fit score by 50 points — the job stays in the DB but won't be shortlisted for drafting.

targets — your company list:

targets:
  - company: Vercel
    domain: vercel.com
    roles: [developer advocate, devrel, developer experience]

roles are the search queries sent to LinkedIn Jobs via Netrows. The agent runs each query and filters results to the named company client-side, so you don't get cross-company noise.

Step 2 — Load your resume

Drop your resume at resume/resume.docx, then run:

node job.mjs resume

This extracts your resume text locally with mammoth, sends it to the AiAS LLM to pull structured facts (accomplishments, companies, skills, links), and caches the result to resume/facts.json. Every subsequent draft call reuses this cache — no re-extraction, no extra API cost.

The facts JSON looks like:

{
  "name": "Mark Allen Evans",
  "current_title": "Founder / AI Platform Engineer",
  "years_experience": 12,
  "companies": ["Interchained LLC", "..."],
  "skills": ["Node.js", "Python", "AI orchestration", "..."],
  "accomplishments": [
    "Built AiAssist Secure — multi-tenant AI orchestration platform serving 40+ orgs",
    "Shipped @aiassist-secure/intelligence-mcp, an npm MCP package with 22 sources",
    "..."
  ],
  "links": { "github": "...", "linkedin": "..." }
}

Step 3 — Scan

node job.mjs scan

For each target company the agent:

  1. Fetches company context — hits the company's main site and blog, strips HTML, summarises with the LLM, writes to company_cache/{company}.json. This snapshot is what makes the draft emails feel current and specific rather than generic. Cached for 24 hours — subsequent scans skip the fetch.

  2. Searches LinkedIn Jobs via Netrows for each role query, then filters client-side to the exact company. You get only real matches, not keyword-adjacent noise.

  3. Scores each job (0–1):

    • Base: 0.5
    • +0.05 per skill keyword hit in title/description
    • +0.10 if the title matches a target_roles entry
    • +0.10 if remote and remote_ok: true
    • +0.15 if posted < 48 h ago, +0.05 if < 1 week
    • −0.50 if any deal_breaker found in title/description

  4. Shortlists jobs with fit_score >= 0.6 and searches LinkedIn People for a relevant team member at that company.

  5. Finds their email via Netrows email finder (5 credits per lookup).

To scan a single company:

node job.mjs scan --only=Vercel

To see raw API responses:

node job.mjs scan --only=Vercel --verbose

Step 4 — Review the list

node job.mjs list

Output:

  NEW   Developer Advocate                         [fit:75%]
        Vercel · Remote · linkedin_jobs

  NEW   Developer Relations Engineer               [fit:70%]
        Groq · Remote · linkedin_jobs

  NEW   Global Trade Compliance Intern             [fit:0%]
        Cloudflare · Lisbon (Hybrid) · linkedin_jobs

Internships with intern in the title score 0% and appear at the bottom. Only fit >= 60% jobs get people search + email lookup + drafting.

Step 5 — Draft

node job.mjs draft

For each shortlisted job with status new, the agent:

  1. Loads your resume facts from cache
  2. Loads the company context snapshot from company_cache/
  3. Calls the AiAS LLM (Groq by default) with:
    • The job title, location, and description
    • The company context — what they're currently building
    • Your best-matching accomplishment
    • The recipient's name and role (if found)
  4. Gets back { subject, body } — under 120 words, practitioner-to-practitioner tone

The result is stored in SQLite with status drafted. Nothing is sent — you control that.

To draft a single job:

node job.mjs draft --only=<job_id>

Step 6 — Export for review (PDF)

node job.mjs report

Generates reports/job_report_{timestamp}.pdf with:

  • Cover page: your name, scan date, summary stats
  • One page per drafted email: job title, company, recipient, subject, body
  • Footer with job URL

Review the PDF. When you're happy with an email, copy the body out and send it manually. Mark it sent:

# (coming soon — node job.mjs send --id=<job_id>)

Archive jobs you're not pursuing:

node job.mjs archive --id=<job_id>

Credit costs (Netrows, €49/mo = 10k credits)

Action Credits
LinkedIn Jobs search (per query) ~5
LinkedIn People search (per job shortlisted) ~5
Email finder (per contact) 5
A full scan of 10 companies, 3 role queries each, 3 shortlisted ~200

A full 10-company scan with people search costs roughly 2% of your monthly credit budget.

Troubleshooting

"Unexpected token '<'" from any source — the endpoint returned HTML (bot detection / Cloudflare challenge). That source is blocked from server environments. The LinkedIn Jobs endpoint via Netrows is the reliable path.

"AiAS chat error 401" — AIAS_API_KEY not set or expired. Company context fetch will fall back to raw text; resume extraction and email drafting will fail.

"0 results" for all companies — check --verbose output. Netrows may return data: null for queries with no results, which is correct (no open roles right now). Try again next week when new roles are posted.

Resume facts not extracting — make sure your file is a real .docx (not .doc or a PDF renamed to .docx). Run node job.mjs resume --verbose to see the raw extraction output.

Environment variables

Variable Required Default Notes
NETROWS_API_KEY Yes — Netrows API key
AIAS_API_KEY For drafts — AiAssist Secure API key
AIAS_API_BASE_URL No https://api.aiassist.net Override for local dev
AIAS_PROVIDER No groq LLM provider (groq, openai, anthropic)
AIAS_MODEL No llama-3.3-70b-versatile Model override