About Xiaofeng Shi

Current work

I am an AI researcher and engineer at BAAI. I build foundation-model systems for specialized domains, covering the full path from data and evaluation to post-training, retrieval, multimodal reasoning, and deployment. My recent work includes mechanical-drawing understanding, enterprise knowledge question answering, and agentic scholarly retrieval.

Career path

My path has taken me through ByteDance and Meituan to BAAI. Across industry engineering and applied research, the recurring goal has been the same: turn broad model capability into dependable tools for real technical settings, while measuring both gains and trade-offs.

Open research

I use open source to make methods inspectable and reusable. When it is responsible to do so, I publish the paper, code, data, model, or evaluation resource. The Research page is a maintained inventory of papers, datasets, models, and software with direct links to the original resources. Published Chinese patent applications are listed separately on Patents.

Collaboration

I am open to research and engineering collaboration around domain post-training, retrieval and agent evaluation, multimodal reasoning, and high-quality open data. For work, email xfshi@baai.ac.cn. For personal correspondence, use xiaofengalg@gmail.com.

Find me on GitHub, Hugging Face, ORCID, and Google Scholar.

中文简介

我目前在北京智源人工智能研究院从事 AI 研究与工程工作，关注行业大模型、后训练、开放数据、检索智能体和多模态推理。我曾在字节跳动和美团工作，并持续记录可公开复核的研究与工程实践。站内的研究页汇总论文、数据、模型与软件，专利页整理已公布的中国专利申请；2018 至 2024 年间发布的中文文章保留在技术档案与归档中。