Demonstrating TOFFEE: A Learned System for Synthesizing Data Agent Trajectories at Scale

2026-07-07

AI Summary

Existing data agents struggle to generalize in heterogeneous enterprise settings, primarily due to the scarcity of high-quality trajectory data tailored to new environments and analytical workflows. To address this challenge, this work proposes TOFFEE, a scalable trajectory synthesis framework that efficiently generates high-fidelity agent trajectories from a given data environment by integrating Monte Carlo tree search, adaptive model selection, and cross-task prefix reuse. The framework incorporates a task pool construction module, a trajectory explorer, and a learned cost model, enabling both supervised fine-tuning and in-context learning. End-to-end application validation demonstrates that the synthesized trajectories substantially enhance agent performance, yielding significant improvements in fine-tuning efficacy and reasoning capabilities on complex, previously unseen data analysis tasks.

Abstract

LLM-powered data agents are playing an increasingly important role in data-driven decision making. However, existing data agents struggle to generalize to unseen data environments and analytical workflows, especially in heterogeneous enterprise settings. This creates a growing need for synthesizing high-quality data agent trajectories that capture complex analytical workflows for given data environments. Such trajectories support two key downstream uses: they can serve as supervised finetuning (SFT) data that adapts data agent models to the target domain, and as in-context learning (ICL) demonstrations to guide general-purpose LLMs in unfamiliar data environments. Thus, we introduce TOFFEE, a system for synthesizing high-quality data agent trajectories from given data environments via Monte Carlo Tree Search (MCTS) with adaptive model selection and cross-task prefix reuse. We show that TOFFEE can effectively generate scalable trajectory data for complex analytical tasks across heterogeneous environments. In this demonstration, we present the system framework of TOFFEE, including its task pool construction, trajectory explorer, and learned cost model. We also introduce the web interface of TOFFEE and its workflow, and demonstrate two end-to-end scenarios: trajectory synthesis for data agent finetuning, and demonstration-augmented data agent reasoning.

Problem

• data agent
• trajectory synthesis
• generalization
• heterogeneous environments
• analytical workflows

Innovation

• Monte Carlo Tree Search
• adaptive model selection
• cross-task prefix reuse
• data agent trajectories
• synthetic data generation