Pain addressed: “We trained something in notebooks but nothing reproducible survives handoff.”

What you receive: a batch codebase with agreed inputs and outputs, model or feature artefacts, manifest or version fingerprints you can grep in logs, and tests or sanity checks tied to milestones.

In scope: explicit paths, preprocessing version, rerun rules, documented failure behaviours. Monitoring or drift reports can be phased add-ons with their own milestones.

Out of scope: unmanaged 24/7 babysitting beyond agreed checkpoints, retraining on live fire without contract, or production incident response unless separately agreed.

Outcome: batch jobs your team or cloud scheduler can run with clear success/fail signals. Portfolio projects show the engineering style; your engagement uses your data and boundaries.