How santifer/career-ops Classifies Jobs into LLMOps, Agentic, PM, SA, FDE, and Transformation Archetypes

Career-Ops uses a constrained prompt-based classification system paired with deterministic JavaScript extraction to automatically categorize every job description into one of six predefined archetypes.

Career-Ops is an open-source evaluation framework that analyzes job postings through a hybrid AI-and-code pipeline. When you submit a job description, the system leverages the oferta skill to classify the role using a closed Domain list that forces the LLM to choose between LLMOps, Agentic, PM (Product Manager), SA (Solutions Architect), FDE (Full-Stack/Dev Engineer), and Transformation archetypes.

The Six Career Archetypes Defined

The classification taxonomy organizes modern tech careers into six distinct buckets based on core responsibilities:

• LLMOps: Roles focused on AI platform operations, observability, evaluation pipelines, and production reliability (e.g., "AI Platform / LLMOps Engineer")
• Agentic: Positions centered on building autonomous agents, tool-use implementation, and advanced prompting strategies
• PM: Jobs involving product vision, roadmap ownership, and stakeholder alignment across technical and business teams
• SA: Architecture-heavy roles responsible for designing end-to-end solutions and integration patterns
• FDE: Generalist engineering positions covering front-end, back-end, and full-stack development responsibilities
• Transformation: Senior leadership or strategy-driven roles that drive organizational change and digital transformation initiatives

Prompt-Driven Classification in modes/oferta.md

The classification originates in the evaluation prompt located at [modes/oferta.md](https://github.com/santifer/career-ops/blob/main/modes/oferta.md). The prompt contains a constrained Domain bullet list that explicitly limits the LLM to the six valid archetypes:

- Domain (platform/agentic/LLMOps/ML/enterprise/backend/frontend/devops)

When the LLM processes a job description, it evaluates the content against this closed list and outputs a standardized markdown line:

**Archetype:** AI Platform / LLMOps Engineer

This deterministic prompting strategy prevents drift by forcing the model to select from predefined buckets rather than generating free-form labels. The same Domain constraint appears across all 14 localized versions of the prompt (e.g., modes/pt/oferta.md, modes/de/angebot.md), ensuring consistent classification regardless of input language.

Regex-Based Extraction in analyze-patterns.mjs

After the LLM generates the report, the system extracts the archetype using deterministic regex matching in analyze-patterns.mjs:

// locate a line that starts with "Archetype"
const archMatch = reportText.match(/^\*\*Archetype\:\*\*\s*(.+)$/m);
if (archMatch && !report.archetype) report.archetype = archMatch[1].trim();

The extraction logic handles multilingual variations (e.g., "Arquetipo" in Spanish) through additional regex patterns around lines 228-240. The extracted string populates report.archetype, which feeds downstream analytics, PDF generation via generate-pdf.mjs, and tracking TSV exports.

Keyword Reinforcement via modes/_shared.md

To improve classification accuracy, the system injects a keyword lookup table from [modes/_shared.md](https://github.com/santifer/career-ops/blob/main/modes/_shared.md) into every evaluation prompt:

| AI Platform / LLMOps | "observability", "evals", "pipelines", "monitoring", "reliability" |
| Platform / LLMOps    | Production systems, observability, evals, reliability |

These keywords provide concrete signals to the LLM. For example, job descriptions mentioning "Langfuse," "Weights & Biases," or "evaluation pipelines" trigger the LLMOps classification because those terms appear in the archetype's keyword list (also referenced in templates/portals.example.yml lines 82-88).

Multi-Language Localization Support

Career-Ops supports 14 language packs, with each locale maintaining its own oferta.md variant:

• Ukrainian: modes/ua/oferta.md (line 18)
• Portuguese: modes/pt/oferta.md (line 16)
• German: modes/de/angebot.md (line 16)

The extraction engine recognizes localized archetype headings, ensuring that a report generated in Spanish containing **Arquetipo:** Arquitecto de Soluciones is correctly parsed and normalized to the canonical SA archetype.

Complete Classification Workflow

The end-to-end archetype assignment follows this deterministic pipeline:

1. Prompt Injection: The user invokes /career-ops {job description}, triggering the oferta skill with the Domain list and keyword table from _shared.md
2. LLM Evaluation: The model analyzes the job description and outputs a markdown report containing the **Archetype:** line
3. Pattern Extraction: analyze-patterns.mjs parses the report using regex to populate report.archetype
4. Data Persistence: The archetype is written to the tracking TSV, rendered as a badge in the PDF output, and fed into analytics engines for conversion-rate analysis
5. User Display: The final report (see [examples/sample-report.md](https://github.com/santifer/career-ops/blob/main/examples/sample-report.md)) presents the archetype alongside scoring breakdowns and recommendations

Summary

• Career-Ops classifies jobs using a constrained Domain list in modes/oferta.md that forces the LLM to choose between six specific archetypes
• The extraction logic in analyze-patterns.mjs uses regex pattern matching (/^\*\*Archetype\:\*\*\s*(.+)$/m) to harvest the classification from markdown output
• Keyword reinforcement in modes/_shared.md improves accuracy by mapping specific terms (e.g., "Langfuse," "agentic") to their respective archetypes
• The system supports 14 languages through localized prompt variants and multilingual regex patterns
• Extracted archetypes drive downstream functionality including PDF badge rendering, analytics tracking, and personalized career recommendations

Frequently Asked Questions

What happens if a job description fits multiple archetypes?

The prompt instructions in modes/oferta.md direct the LLM to select the single most appropriate category based on primary responsibilities. If a role spans multiple domains (e.g., a PM with LLMOps exposure), the keyword reinforcement table in _shared.md guides the model toward the dominant function.

Can I customize the archetypes or add new categories?

Yes. You can modify the Domain list in your language's oferta.md file and update the keyword table in modes/_shared.md to include new archetypes or terminology. Changes take effect immediately on the next evaluation without requiring code changes, though you should verify that analyze-patterns.mjs can extract your new archetype labels.

How does the system handle non-English job descriptions?

The classification supports 14 languages through localized prompt files (e.g., modes/pt/oferta.md, modes/de/angebot.md). The extraction logic in analyze-patterns.mjs recognizes translated archetype headings like "Arquetipo" or "Archétype," ensuring consistent categorization across language barriers.

Where can I see the archetype classification in the final output?

The archetype appears in three locations: as a **Archetype:** heading in the markdown report (examples/sample-report.md), as a visual badge in the generated PDF via generate-pdf.mjs, and as a dedicated column in the tracking TSV file used for analytics and conversion-rate monitoring.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}