01 / AI for science · Scientific software

TritonDFT

Can an AI system manage a complete density-functional-theory workflow while keeping every scientific decision inspectable?

TOC graphic for TritonDFT: agentic workflows for first-principles simulation

Live platform · 2026

Software platform

Overview

TritonDFT is a multi-agent framework that turns a natural-language materials question into an end-to-end density-functional-theory workflow. From a browser, it plans calculations, generates Quantum ESPRESSO inputs, runs HPC jobs, checks convergence, analyzes properties, and refines the accuracy–cost trade-off.

Key resultA single browser interface coordinates the complete Plan–Execute–Refine loop while keeping the scientific workflow inspectable.

My contributionMulti-agent DFT workflow development and materials-science validation.

Can an AI system manage a complete density-functional-theory workflow while keeping every scientific decision inspectable?

Approach

  • A four-agent Planner, Executor, Analyzer, and Refiner architecture
  • Natural-language planning and Quantum ESPRESSO input generation
  • HPC execution, output parsing, validation, and convergence checks
  • Pareto refinement of cutoffs, k-point grids, functionals, accuracy, and cost

What emerged

  • Specialized agents can coordinate the full Plan–Execute–Refine loop through one browser interface.
  • TritonDFT reduces repetitive steps that ordinarily take researchers minutes or hours to seconds or minutes.
  • DFTBench evaluates scientific expertise, HPC knowledge, optimization trade-offs, and cost efficiency.
  • The platform produces inspectable workflows and publication-ready results while preserving scientific control.

Try TritonDFT

Use the live interface here, or open it in a new tab for the full workspace.

Open full application