Tapilot-Crossing is a benchmark dataset for evaluating Language Model agent performance on interactive data analysis tasks. It contains 1,094 user intents across 946 entries, organized into 8 categories spanning 5 data domains like credit card risk and tennis. The dataset was created by birdsql and last updated on Hugging Face in April 2026.
Use Cases
- Benchmarking LLM agent performance based on simulated user-agent interactions.
- Evaluating agent reasoning across diverse data domains like credit card risk and nutrition.
- Training agents for interactive data analysis tasks based on structured user intents.
- Studying cost-effective simulation methods for human-agent dialogue.
Strengths
- Contains 1,094 distinct user intents for agent evaluation.
- Covers 5 diverse data domains, including credit card risk and ATP tennis.
- Organizes entries into 8 categories for structured analysis.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown, which may limit suitability assessment.
- Description metadata is limited; actual data quality requires manual inspection after download.
Provenance
- Source
- birdsql on Hugging Face.
- Collection Method
- Likely constructed as a benchmark for simulating user-agent interactions via DECISION COMPANY.
- Time Range
- null
- Freshness
- Last updated 2026-04 17:20:21; freshness should be verified.
- Geography
- null