ViseGPT: Towards Better Alignment of LLM-generated Data Wrangling Scripts and User Prompts
summary
ViseGPT extracts eight types of constraints from natural-language prompts, generates targeted test cases, runs LLM-produced data-wrangling scripts against them, and provides a Gantt chart mapped to script lines plus a script-refiner that turns failing tests into targeted LLM prompts. A formative study (N=8) shapes the design; a user study (N=18) shows significant reductions in debugging time and higher issue detection and correction rates. Results indicate practical, iterative improvement of generated scripts. N=18 user study
AI-written summary · about
chapters (9)
- 0:00Introduction to ViseGPT
- 1:29Limitations of existing methods
- 2:54Key challenges in constraint-based testing
- 3:36System workflow overview
- 5:00Interactive Gantt chart visualization
- 6:06User interactions in ViseGPT
- 7:33Usability evaluation results
- 8:38Contributions and future work
- 9:00Q&A session
related talks
48mBeyond the Prompt: Evaluating, Testing, and Securing LLM Applications
01h 15m10x Development: LLMs For the working Programmer
41mJudge the Judge: Building LLM Evaluators That Actually Work with GEPA
39mBeyond the Prompt: Evaluating, Testing, and Securing LLM Applications
35mEvaluating LLM systems: the nuts and bolts
50mBeyond the Prompt: Evaluating, Testing, and Securing LLM Applications
46mBeyond Basic Prompting: Supercharging Open Source LLMs with LMQL's Structured Generation
38mMaking LLMs reliable - A practical framework