Skip to content

Personalkostenkalkulator

Web-based application for calculating personnel costs in research and third-party funded projects under the TV-L collective agreement. Input can be provided either in natural language (chat) or through a structured form. An LLM-assisted component handles understanding the request and extracts the calculation parameters; the calculation itself is performed deterministically without LLM involvement. Results are displayed in tables and can be exported as Excel or PDF files, including a breakdown of employer contributions in the BUND application format.

At a glance

  • Describe personnel costs for third-party funded projects in natural language and have them fully calculated
  • Combine multiple positions with different pay grades, levels, and durations within a single project
  • Account for level progressions, year-specific pay scale increases, and multi-year project timelines
  • Calculate student assistants on an hourly basis
  • Export results as an Excel file including a BUND format with itemised employer contributions
  • Ask follow-up questions about the completed calculation and have individual positions or yearly slices explained
  • Be guided to correct input through clarification questions or a full form when input is ambiguous

Highlights

Unlike a direct LLM prompt or a plain spreadsheet, the application complements understanding of the request with a deterministic calculation and ensures that the values produced are reproducible and traceable.

  • Strict separation of understanding and calculation — The LLM extracts only parameters from the input; the calculation itself is rule-based and runs without AI involvement. As a result, outputs are exact and independent of the model used.
  • Hybrid input — Natural-language input and a structured form are available in parallel and can be combined within the same session. Targeted clarification questions are issued when input is incomplete.
  • Multi-stage validation — The LLM response is checked against a schema; missing required fields are automatically converted into a clarification question, and repeated parse failures trigger a fall-back to the manual form.
  • Year-specific pay scale increases — Increase rates can be set individually per calendar year or disabled entirely; multiple forecast scenarios can be evaluated without re-entering positions.
  • Itemised employer contributions — For BUND applications, pension, health, unemployment, and long-term care insurance are reported separately.
  • Q&A on the existing calculation — Follow-up questions such as "What does the project cost in 2027?" or "Which position is the most expensive?" are answered from values already computed, not by recalculating through the LLM.
  • Traceable result structure — Each position is broken down into yearly slices with base salary, employer-gross, special payment, and total cost; planned level progressions are reflected in the year in which they take effect.
  • OpenAI-compatible LLM endpoint — Connection to any OpenAI-compatible chat API; local servers such as LM Studio, Ollama, or vLLM are supported without modification.
  • Importable pay scale data — The pay scale table, special payment rates, and hourly rates can be updated from an Excel file via a command-line tool.