Archive¶
Note
These pages document the original experiments and are no longer actively maintained. Current tool descriptions can be found under Tools.
LLM-Supported Software Development: Methodological Insights from Practice¶
A Series of Experiments to Explore New Development Approaches¶
Since the availability of powerful Large Language Models, the question arises: How is the software development process changing? What are the practical possibilities - and where are the limits? This series documents methodological observations from various experiments and development projects realised with LLM support.
As a Comic¶
Overview of AI Tools (as of experiments)¶
| Tool | Function | Key Challenges |
|---|---|---|
| Chart Tool | Creates interactive diagrams from CSV/Excel through natural language input | Automatic error correction in generated code; intelligent intent recognition |
| Chat | Web-based interface for dialogues with locally operated LLMs | Session management; automatic summarization |
| Diagram Tool | Generates technical diagrams from natural language | Multi-stage validation pipeline; Mermaid syntax corrections |
| Doc | Automatic summarization of PDF/DOCX/ODT documents | Synthesis of partial summaries for extensive documents |
| Flex Mapping | Generic information extraction system for websites | Two-phase extraction with QA layer; entity normalization |
| AI Survey | Intelligent survey system with automatic follow-up questions | Clarity score evaluation; adaptive follow-up generation |
| PPT Helper | Develops presentation structures from uploaded documents | Multiple agents; iterative refinement |
| STT Helper | Transforms machine transcripts into professional documents | Multi-stage LLM workflows; context-based technical term recognition |
| Staff Cost Tool | Calculates personnel costs for funded projects | Extraction of pay grades from free text; annual slice calculation |
| Style Tool | Transforms texts according to predefined style profiles | Extraction and application of linguistic features |
| Stylish | Slider-based text style editor with 34 sliders | Intensity levels; balance between control and usability |
| TextTool | Iterative text editing with predefined functions | History management; balance between standard and customization |
| Translate | Translates documents while preserving structure | Intelligent chunking; parallel processing; glossary management |
| TalkToDocuments | Dialogue-based document analysis with source references | Management of many documents; automatic content preparation |
| Web Helper | Migrates web content from Plone to TYPO3 | JSON structure recognition; LLM-supported content analysis |
The Experimental Approach¶
The documented projects include tools of varying complexity: from document processing systems and translation pipelines to code analysis tools. Development times ranged from one to seven hours for functional prototypes.
Key Findings¶
Specification as a success factor: Across all projects, the quality of the specification proved crucial. The clearer the functional requirements, technical dependencies and architectural decisions were defined upfront, the more error-free the implementation was.
Active control required: LLMs tend to favour complex solutions. Consistently demanding simple approaches (KISS principle) proved to be a necessary control task.
Structuring for maintainability: Clear modularisation with size limits per file improved both code quality and maintainability by LLMs.
Changed developer role: Requirement clarification, architectural decisions and critical evaluation of LLM proposals are becoming core competencies.
Detailed experiment documentation can be found in the subsections Experiments, Usage and Results.