Sep 10 · Git + Linux/Shell Toolkit
Branches, rebase, PR hygiene — plus the shell commands that buy 80% of the leverage.
13 live 2-hour classes. One real production-shaped ML service shipped to your GitHub by Demo Day.
You build the same ML service all cohort long, adding one production layer per class. Thursdays, 2 hours live. Here's what each week covers.
Branches, rebase, PR hygiene — plus the shell commands that buy 80% of the leverage.
uv, FastAPI skeleton, pytest. Code shape that survives review, not notebook patterns.
The real model goes in. Loading, versioning, contract design, failure modes.
Multi-stage builds, layer-cache discipline, GHCR. ML images shouldn't be 4 GB.
Pods, Deployments, Services, HPA. The mental model — not the certification.
GitHub Actions. PR runs tests, main builds and pushes. OIDC over PATs.
Prometheus, Grafana, RED metrics — and the drift signals traditional APM misses.
Structured logs, correlation IDs, OTel tracing. One ID threaded end-to-end.
Postgres, indexes, EXPLAIN, transactions, connection pooling.
Airflow DAG that retrains weekly. MLflow tracking. Idempotency patterns.
Terraform for ECR + EKS + RDS. IAM the way it's actually written. Cost levers.
Trace one request through every layer. Capstone PR opens. What the role really looks like.
vLLM, batching, cost economics — then the cohort presents their capstones. Graduation.
No class Nov 26 (Thanksgiving). Recordings posted within 24 hours if you miss a week.
The cohort is capped at ~30 seats. I read every application personally. You'll hear back within 3 business days with the payment link if accepted.