JobFoundry Pipeline Architecture

JobFoundry is structured as a decoupled, multi-stage processing pipeline that converts unstructured web job listings into deduplicated, score-ranked opportunities with tailored ATS resumes.

Interactive Pipeline Visualizer

Click through each stage to view the architectural rationale and code implementation:

// extension/content-script.js
const payload = {
  url: window.location.href,
  title: extractField('title'),
  company: extractField('company'),
  descriptionHtml: extractField('description'),
  extractedAt: new Date().toISOString()
};
// Sent directly to local JobFoundry daemon

// server/ingest/dedup.js
const hashA = computeSimHash(jobA.description);
const hashB = computeSimHash(jobB.description);
const distance = hammingDistance(hashA, hashB);

if (distance <= 3) {
  // Mark as duplicate of canonical job
  return deduplicate(jobB, jobA.id);
}

{
  "fitScore": 88,
  "matchGrade": "A",
  "keyStrengths": ["Distributed Systems", "TypeScript", "PostgreSQL"],
  "gapAnalysis": ["Limited Go experience mentioned in posting"],
  "recommendation": "Strong apply"
}

// server/tailor/engine.js
const tailoredResume = await generateTailoredProfile({
  masterProfile,
  targetJob,
  template: 'ats-modern'
});
await compileTypst(tailoredResume, 'output.pdf');

GET /api/jobs?status=scored&minScore=75
Response: 200 OK
[
  { "id": "job_9412", "title": "Staff Engineer", "score": 94, "stage": "READY_TO_APPLY" },
  { "id": "job_8204", "title": "Backend Lead", "score": 88, "stage": "APPLIED" }
]

Stage 1: Browser Ingestion

The companion browser extension (extension/) injects lightweight content scripts when visiting supported job boards and ATS domains.

• Extraction Strategy: Prefers structured JSON-LD (schema.org/JobPosting) when present. Falls back to curated CSS selector trees for dynamic Single Page Applications (SPAs) like Workday or Ashby.
• Payload Delivery: Sends parsed data via POST /api/jobs/ingest to the local daemon, accompanied by an authentication token generated during initial pairing.

Stage 2: SimHash & Deduplication Engine

Job postings are frequently syndicated across multiple aggregators (e.g., posted on Greenhouse, scraped by LinkedIn, and mirrored on Indeed). JobFoundry avoids duplicate scoring through a 64-bit SimHash locality-sensitive hashing algorithm:

1. Tokenization & Normalization: Strips boilerplate disclosures (EEOC, copyright), converts text to lowercase, and extracts weighted n-grams.
2. 64-bit Hash Fingerprint: Computes a compact bitvector where similar documents have small Hamming distances.
3. Hamming Distance Clustering: If distance(hashA, hashB) <= 3, the job is identified as a cross-posting. The canonical posting is maintained while appending the new source URL as an alternative link.

Stage 3: LLM Fit Scoring & Criteria Evaluation

The scoring engine (server/scorer/) compares your master candidate profile against the job description using local LLMs (such as Llama 3 via Ollama or vLLM) or remote OpenAI-compatible endpoints.

• Strict JSON Schema: Responses must satisfy an Ajv-validated schema ensuring valid integer scores (0–100), key matched qualifications, identified gaps, and actionable interview talking points.
• Zero Hallucination: The model evaluates fit against provided profile facts only. It does not invent or assume credentials not present in your profile.

Stage 4: Resume Tailoring & ATS Export

Rather than rewriting your entire career history, the tailoring engine (server/tailor/) re-ranks and highlights your genuine experiences that directly align with the job's stated requirements.

• Typst Compilation: Compiles clean, modern resumes using Typst, producing pixel-perfect PDF and plain-text ATS formats.
• Deterministic Output: Generates verifiable, professional documents ready for submission.

Stage 5: Local Kanban Dashboard

The web interface (server/web/) provides a real-time Kanban board organized by pipeline status:

• DISCOVERED: Newly ingested jobs awaiting automated processing.
• SCORED: Jobs evaluated with fit scores and match grades.
• READY_TO_APPLY: High-scoring jobs with generated tailored resumes.
• APPLIED: Active applications with date tracking.
• INTERVIEWING: In-progress interview stages and notes.