PoC tasks (high level)

Prepare infra: provision 3 VMs (control VM for Dagit+daemon, Postgres VM, 1 worker VM). Ensure networking between them and backups for Postgres.
Install dependencies: OS updates, Python 3.9+/virtualenv, Dagster, dagit, psycopg2, and any SQL Server drivers (ODBC / pyodbc).
Postgres setup: create Dagster metadata DB, configure backups and retention (20 days), set resource limits.
Dagster baseline: create a simple repository with one pipeline (job) that reads from SQL Server, performs a small Python transformation, and writes to a SQL Server/warehouse table.
Authentication: validate connectivity to SQL Server from the worker VM (use Kerberos/AD or SQL auth per your environment).
Scheduling & runs: configure schedule sensor or cron to trigger the job; run several executions and observe logs/metadata in Dagit/Postgres.
Logging & monitoring: ensure logs are persisted, accessible from Dagit, and rotated/archived per retention policy.
Power BI integration: verify transformed data is queryable from Power BI (DirectQuery or scheduled import) against SQL Server.
Backup & restore test: restore Postgres metadata once from backup to confirm backups are usable.

Acceptance criteria (what 'done' looks like)

Dagit UI is accessible from your network and shows repository + runs.
Dagster runs complete successfully on the worker for 3 consecutive scheduled runs (no manual intervention).
Metadata for runs (run history, logs) stored in Postgres and retained for 20 days.
Worker VM can connect to SQL Server using your chosen auth method and can read/write target tables.
Power BI can query the transformed tables and render a simple report.
Postgres backup + restore verified (test restore within retention window).
Basic failure handling: when the job fails, Dagster shows error and logs, and retries (if configured) behave as expected.

Quick risks & mitigations

Risk: AD/Kerberos complexity — Mitigation: test SQL auth first, then enable Kerberos once PoC works.
Risk: Network/firewall blocks — Mitigation: validate ports (Postgres, Dagit) and IP allowlists early.
Risk: Director scrutiny over VM count — Mitigation: show minimal VM approach and emphasize temporary nature of PoC and rollback plan.

Ready to receive

Do you want me to now (choose one):

Generate the ready-to-send infra request for your infra team (VM specs, networking, backups, AD notes)
Generate a short runbook/playbook for installing Dagster, configuring Postgres, and a sample pipeline
Generate both infra request and runbook

Optional: one short sentence if you have a preference