DataPilot AI: An Autonomous Multi-Agent Data Science Platform for Automated Analytics, Model Engineering, and Explainable Insights
Translating raw enterprise data into robust predictive models remains bottlenecked by the labor-intensive nature of exploratory analysis, data sanitization, leakage-free feature engineering, and model tuning. In this paper, we present DataPilot AI, an autonomous data science platform built on a collaborative multi-agent architecture powered by Large Language Models (LLMs). DataPilot AI coordinates specialized agents for data profiling, anomaly cleaning, automated feature transformation, Bayesian hyperparameter optimization, and explainable AI diagnostics using Shapley Additive Explanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME). To prevent silent failures, the platform incorporates strict fold-wise leakage prevention, deterministic static analysis guards, and an isolated, containerized execution runtime. On standard tabular benchmarks, DataPilot AI is designed to evaluate predictive performance and workflow efficiency against traditional AutoML baselines. The proposed framework aims to reduce pipeline execution latency and automate the process of transforming raw datasets into validated, explainable machine learning pipelines.
Authors
- Akash Karale (ORCID: https://orcid.org/0009-0007-3519-4553)
- Megha Wankhade
- Chetan Thorat
- Prashant Rode
- Shubham Shidore
Institutions
- MIT Art, Design and Technology University (IN)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-17
- DOI
- https://doi.org/10.5281/zenodo.24297138
- Primary Topic
- Machine Learning and Data Classification
- Type
- preprint