Do a bit.
ProductsWritingContact

Research.

I work at the intersection of HCI, visualization, and AI agents — using interaction and visualization to help people communicate information more clearly, and more beautifully.

h-index 5 · 206 citations · Google Scholar

  • MixLoRA

    Fine-tuning LLMs more efficiently by routing through a mixture of LoRA experts.

    arXiv 2024· cited 161

  • DataLab

    A unified platform that lets non-technical users query and visualize their data through an LLM.

    ICDE 2025· cited 15

  • DLRover-RM

    Cutting cloud cost for deep recommendation model training by reusing resources across jobs.

    arXiv 2023· cited 14

  • Multimodal Prompt for Visualization

    Authoring data visualizations by sketching and pointing — multimodal prompts for LLMs.

    arXiv 2025· cited 9

  • VIS-Shepherd

    A critic model that catches mistakes in LLM-generated data visualizations.

    arXiv 2025· cited 5

  • IGenBenchlead

    Benchmarking how reliably text-to-infographic models reproduce the data they're given.

    arXiv 2026· cited 2

  • ViviDoclead

    Generating interactive documents through human-agent collaboration.

    arXiv 2026

← back home

Do a bit.
AI-native studio

An AI-native studio. Building quiet tools and writing about a more deliberate way to work and live.

ExploreProductsWritingContact
© 2026 · doabit.devjust do a bit, every day.