A notebook becomes a Lambda

The analysts had the algorithm and they lived in notebooks. Instead of translating each change into application code, a converter turns the notebook itself into the Lambda handler: first a local script, then a deployed stage, then a rewrite of the converter once the pattern proved itself.

What the notebook had to give up: local files, printed results, cells that only work in order. What the platform had to add: an input contract, an output contract, and canaries that exercise the handler without writing anything. The analyst never opens a pull request; the pipeline never runs untested code.

flowchart LR
  N[Notebook in S3] --> C[Converter] --> H[Lambda handler] --> A[API stage]
  K[Canary, dry run] --> A