CS-003 / Logistics · enterprise / Embedded team
Stood up an in-house ML team and walked away when they were ready.
The internal team had built two models that worked in notebooks but had never made it to production. They needed a reference for what 'ready' looks like — eval surface, deployment patterns, on-call.
// problem
The internal team had built two models that worked in notebooks but had never made it to production. They needed a reference for what 'ready' looks like — eval surface, deployment patterns, on-call.
// constraints
- We were not hiring or managing the in-house team
- Goal explicit: leave them stronger, not dependent
- Quarterly review with their head of engineering on whether to extend
// approach
What changed.
Embed
One ML lead at 0.8 FTE, one platform engineer at 0.5 FTE. Same Slack, same standups.
Stewarded eval
Set up the eval surface for both legacy models; refused to let new models ship without passing it.
Architecture
Sign-off on every PR for the first three months; advisory after that.
Exit
Three-month wind-down. Ran the last deploy as observers. Final exit review documented strengths, fragilities, and the path to staffing the gap we left.
// results
Measured outcomes.
Models in production
+ 4
0 → 4
Deploy frequency
—
— → Weekly
Time from idea → shadow
—
— → 12 days
In-house engineers shipping solo
Same team, more capable
4 → 4
The hardest thing about hiring senior consultants is getting them to leave. They wrote their own exit criteria and stuck to them.
Head of Engineering
The internal team had built two models that worked in notebooks but had never made it to production. They needed a reference for what 'ready' looks like — eval surface, deployment patterns, on-call.
An embed engagement should have the same kind of exit criteria as a build. Otherwise the relationship calcifies into staff augmentation and stops creating leverage.