一人工程 · solus opus

← 全部作品

一人工程machine-learningsolomlops

solo engineer 没有 machine learning,我自己就是 machine learning

团队有 machine learning。

Machine learning:机器学习。Deep learning:深度学习。Model:模型。Training:训练。Inference:推理。MLOps:ML operations。MLflow:open source MLOps。Kubeflow:Kubernetes ML。$100-$10000 / 月 in ML infrastructure.

每 startup 1 个 ML team + 1 个 MLOps engineer + 1 个 data scientist。$300K-$700K / 年.

solo engineer 没有 machine learning。

我自己就是 machine learning。

我想 train 就 train、想 infer 就 infer、想 deploy 就 deploy、想 evaluate 就 evaluate。我不需要 machine learning、不需要 deep learning、不需要 model、不需要 training、不需要 inference、不需要 MLOps、不需要 MLflow、不需要 Kubeflow.

我没 machine learning 的必要——访问量 1000 不需要 ML。我没 deep learning 的必要——没 DL。我没 model 的必要——commit log 是 model。我没 training 的必要——没 training。我没 inference 的必要——没 inference。我没 MLOps 的必要——没 MLOps。我没 MLflow 的必要——没 MLflow。我没 Kubeflow 的必要——没 Kubeflow。我没 $100-$10000 / 月 的必要——Vercel free + commit log。我没 data scientist 的必要——没 data scientist。我没 $300K-$700K / 年 的必要——没 ML infrastructure.

团队的 machine learning 价值是「让 model 自动学习 pattern + 预测 + 决策」。前提是有 big data、需要预测、需要自动化。solo engineer 没这个问题——访问量 1000、commit log 是 pattern、commit = 预测.

一人工程的 machine learning = Vercel Edge Function + git commit。不是 MLflow + Kubeflow $10000/月 + data scientist team.