AI UNDERDOGSDAILY PICK
AI UNDERDOGS
告别手抄设备台账
End manual asset logging
ljp-777/DeviceLens
对着机柜拍一张照，设备清单、拓扑图
BOM 表，一步到位全给你生成好
Snap one rack photo. It auto-generates your asset
list, network topology, and BOM table — all
structured, all in one step
一拍，出清单
Photo to inventory
运维同学巡检机房
一个机柜二三十台设备
以前手抄一下午还容易出错
DeviceLens 用视觉模型识别照片里的每台设备
直接生成 Markdown 归档文档
Field engineers photograph racks with 20-30 devices. Used
to take an afternoon to hand-transcribe, still riddled
with errors. DeviceLens uses a vision LLM to
identify every unit and generate a structured Markdown
record
★ SIGNAL 1
不止识别，还出文档
Detection is table stakes
很多工具能识别设备就完了
但 DeviceLens 直接输出结构化文档——装备清单
网络拓扑、BOM 表
甚至完整 Markdown
从照片到可归档记录，不让你多走一步
Most tools stop at detection. DeviceLens goes further
— equipment list, network topology, BOM table, and
a full Markdown doc. Photo to archived record
no extra steps
★ SIGNAL 2
30 秒部署
30-second deploy
作者接了火山引擎和 OpenAI 两套视觉后端
国内用户不用折腾 API key
本地部署就是 clone 加配个 .env
README 里连 demo GIF 都放好了
一看就是自己踩完坑写的
Author hooked up both Volcano Engine and OpenAI
as vision backends. Domestic users skip the API
key hassle. Local deploy is just clone plus
configure .env — even the demo gif is
in the README, clearly written after fighting through
the setup themselves
作者原话：「Cursor 订阅快到期了
就用它做了个好玩的项目」
没有宏大叙事
就是觉得这个场景值得做
顺手把视觉识别和结构化文档一起搞定了
Author quote: 'My Cursor subscription is about to
expire, so I used it to build a
fun project.' No grand vision — just saw
a real pain point and decided to fix
it with vision AI
AI UNDERDOGS
机房设备档案，一张照片搞定
One photo generates your rack docs
ljp-777/DeviceLens
关注 · 每天发现更多 AI 神作
github.com/ljp-777/DeviceLens