teLLMe Why (Ain't Nothing but a Jam): Exploratory Causal Analysis of Urban Driving Data

TL;DR - teLLMe is a system that pairs LLMs with formal causal-inference tooling to let analysts ask natural-language causal questions ("How would rain change traffic density?") over observational urban-driving datasets. It matters because it makes rigorous causal exploration accessible while keeping assumptions and uncertainty explicit.

• Builds a structured event table from dashcam annotations, then applies the PC algorithm for causal structure learning with bootstrap-based stability checks.
• A schema-aware LLM maps free-form questions into structured causal queries (treatments, outcomes, subpopulations); effects are estimated via linear regression and DoWhy.
• Outputs a "Causal Card" summarizing effect estimates, adjustment sets, DAG support, and assumptions, plus a short natural-language explanation.
• Validated on BDD-derived traffic events (weather, peak hours, density); framed explicitly as a hypothesis-generation/expert-reasoning aid, not a source of definitive causal claims.