Hi，我是文康丞，一个在医学院苦苦挣扎的医学生 I'm Kangcheng Wen, a medical student trying hard to survive med school.
我在这个网站上分享我在生活和技术方面的所思所想 This site is where I share my thoughts and research on technology, healthcare, and life.
页面导览
NAVIGATION
如果你想看点稍微成型的东西，微研究是一些英文综述和实验的报告，涵盖一切我感兴趣、想研究的领域；如果只是好奇我最近在捣鼓什么、踩了哪些坑， 博客记的都是过程；如果你只是恰好路过、没有什么特别的目的，或者想看一些不那么严肃的东西， 随笔那边什么都有，也可能什么都不是。
关于我和关于这个网站提供了更多背景--想看看这个网站是怎么搭起来的、为什么要搭，或者想知道我是谁，这两个页面都能满足你。
你可以通过顶部的导航栏自由方便地浏览不同部分。
If you want something a bit more polished, the Mini Lab holds some reviews and small experiments - spanning whatever I've been curious about lately. If you're just wondering what I've been tinkering with, the Blog is where the process gets tackled down. If you're only here by accident, or fancy something less serious, Thoughts could be anything - or nothing in particular.
About Me and About This Site fill in the rest - how this place came together, why I bothered, and who's behind it.
Or just use the menu up top if you'd rather skip all that.
此时此刻
Now
最近我在做医疗隐私数据管理、运动医学诊疗智能体等方面的工作。数据管理部分的核心目的是保证隐私数据管理足够安全、使用足够方便，配套工具使用保证数据使用流程完全自动被记录，保证流程完整可复现（这尤其对需要开发数十个小模型的智能体项目有益）；智能体方面正在做一些影像小模型的开发和测评。
谈及智能体，目前主流的通用LLM（Claude、ChatGPT等）已经具有相当的疾病诊断能力，调用他们的API可以方便地做想做的智能体相关的实验（我大概率也没精力和算力去微调它们）。最近我在学习通用智能体的搭建框架LangGraph，暂时的目标是复现由Google Research & Google Deepmind团队研究的AMIE项目，如果你感兴趣，欢迎来和我交流。
工作之外，我最近新入坑了《明日方舟》，也刚开始玩《底特律：变人》，欢迎你给我介绍有意思的游戏或者一起玩。
更新于：
Lately I've been splitting time between medical privacy data management and an AI agent for healthcare. On the data side, I'm trying to keep private data both properly secure and easy to actually export, with additional tools logging everything automatically, so the whole pipeline stays reproducible. On the agent side, I'm working on building and evaluating some visual models for specific medical tasks.
Speaking of agents: I've been learning LangGraph and general agent-related techniques. I'm expected to work my way toward reproducing AMIE, a general medical agent framework by Google Research and DeepMind. Besides, general-purpose LLMs (Claude, ChatGPT, that sort) are already decent enough at diagnosis of common diseases, so running experiments mostly just means calling APIs, which is good and convinient, because fine-tuning one myself is out of reach, compute- and skill-wise. Happy to talk if you're into this stuff.
Outside of work, I've fallen into Arknights and just started Detroit: Become Human. Recommendations (or a co-op partner) are welcome.
Updated:
联系我
Connect
欢迎通过
You can reach me through
怕什么真理无穷，进一寸有一寸的欢喜。
Ever tried. Ever failed. No matter. Try again. Fail again. Fail better.
Thanks for reading