01 — Problem
Fresh graduates usually sign their first PKWT within days, with no way of telling which clauses break the rules: a probation period in a fixed-term contract, a wage under the regional minimum, a flat resignation fine. Unemployment for 15-24 year olds sits at 16.89%, over three times the national rate, so the bargaining power is all on the employer's side. A general chatbot does not help — it cites articles that were amended or struck down, with nothing the user can check.
02 — Approach
One Langflow flow serves three doors: IBM Bob over MCP, the public web app over the Langflow API, and CI. Personal data (NIK, NPWP, phone, bank number, name and address lines) is masked before any text reaches the LLM, and the contract is never written to the vector store. The agent extracts key clauses, retrieves articles and 2026 regional minimum wages from a 64-entry regulation corpus where every article carries its legal status and regime, and reports each finding with the article, its status, a severity and a polite question for HR. A daily bot reads JDIH and Constitutional Court rulings, rejects any text not found word for word in the official PDF, and opens a pull request that CI will not merge until a human reviews it.