Biography

Research shaped by building real systems.

I am Xiaofeng Shi, an AI researcher and engineer at the Beijing Academy of Artificial Intelligence (BAAI). I previously worked at ByteDance and Meituan.

我的研究关注如何让大模型真正进入专业领域，并尽可能开放论文对应的代码、模型、数据与评测资源。

Selected research Technical notes
Portrait of Xiaofeng Shi

Current focus

From data to domain capability

My work connects data construction, post-training, evaluation, and deployable agentic or multimodal systems.

Domain LLMs

Data & post-training

High-quality corpora, supervised fine-tuning, reinforcement learning, knowledge adaptation, and capability retention.

AI agents

Retrieval & knowledge work

Scholarly retrieval, survey generation, agentic RAG, and long-horizon knowledge workflows.

Multimodal reasoning

Industrial visual language

Technical drawing understanding, cross-chart reasoning, visual-language models, and OCR.

Research path

Computer vision to foundation models

My earlier work covered OCR, visual representation learning, deep metric learning, image retrieval, and multimodal representation learning. At BAAI, my focus moved toward open domain data and models, post-training, agentic research systems, and industrial multimodal reasoning.

I welcome research discussions, open-source collaboration, and responsible adoption of the released resources.

Verified profiles