Analytics Engineer
dbt PRs with lineage diffs, impact before merge, and tests from the manifest.
Department · IT & Data
From Slack archaeology to
blast radius in minutes.
Platform, analytics engineering, ML, and governance share one graph: DAGs, models, owners, and incidents.
Overview
IT can say yes without losing the graph.
A rename in staging should not be a 2am mystery. On-call should not open five tabs to learn which dashboards go stale. New hires should not spend three weeks reconstructing tribal wikis.
Company Brain is the control plane above the stack you already run. Sources, transforms, and consumers are named. A breaking change has an owner before it hits a dashboard. Agents only see what the graph allows.
This page is for platform, analytics engineering, ML, BI, and governance. The same map the AE uses on a Tuesday is the map on-call uses at 2am.
- Analytics engineers shipping dbt with impact on the PR
- Platform and on-call tracing DAGs to dashboards
- Heads of data who need coverage and ROI they can show
- Governance leads who need policy that travels with the asset
Before mergeBlast radius on the PR, not after production
< 2hTypical path from alert to cited impact list
One graphCatalog, incidents, and agents share nodes
Company Brain
What Company Brain is, and how it helps IT.
Company Brain is the control plane above the stack you already run. Sources, transforms, consumers, and incidents share one graph. Agents retrieve from that graph before they speak.
On-call stops opening five tabs. A rename in staging shows blast radius before merge. New hires inherit a living catalog, not a wiki.
- Impact on the PR: dashboards, features, and finance reports named before merge.
- When a DAG retries, the first answer is blast radius, not archaeology.
- Access and AI share the same grants. Provisioning is a graph change.
- Unused extracts become visible so you can retire them on purpose.
Use cases
Go from five tabs at 2am to one incident map
dbt & warehouse
Ship PRs without breaking production
Rename a column and see dashboards, features, and finance reports you never knew existed, before merge.
- Impact query in the stack-aware editor.
- Lineage diff on the PR. Tests suggested from the manifest.
- Owners notified when an exec-facing model is in the blast radius.
On-call
Know which dashboards go stale
When extract_charges retries, the first question is blast radius. Not who wrote the model two years ago.
- Airflow and Dagster tasks linked to dbt and BI.
- Paste a cited impact list into Slack in minutes.
- Alert the owner, not the whole company.
Ownership
Onboard in days, not tribal wikis
Schemas, freshness, and owners stay current as the stack changes. If it is not in the inventory, it is not in the brain.
- No more spreadsheet archaeology.
- Governance tags (PII, exec-facing) travel with the asset.
- Unused paths become visible so you can retire them on purpose.
Training paths
Trace features to source tables
Drift and retrains sit on the same graph as the warehouse, so model risk can defend provenance.
- Feature DAG from raw to score.
- Lineage diffs between model versions.
- Which production models consume a column, answered with citations.
Control
Who can see what, in one place
Roles, datasets, and tools share one access story. Provisioning is a change to the graph, not a week of tickets.
- Named roles and dataset grants.
- Agents inherit the same RBAC as humans.
- Fewer tickets. Fewer tribal exceptions.
FinOps
Spend you can attribute
Warehouse, tools, and model runs sit against a team and a product. FinOps stops being a quarterly surprise.
- Attributed spend with a team and a product.
- Unused models and extracts visible enough to kill.
- No surprise bill with no owner.
How it lands
Platform and analytics. No more handoffs.
Metadata-first
Crawl the stack you already run
Snowflake, dbt, Airflow, Kafka. Metadata only. Credentials stay in your VPC.
- Docker Compose in minutes for a first look.
- No production rows leave the network.
- Inventory is the door. Name every source before you add another.
Interface
Consumers get a contract, not a table dump
Grain, freshness, and owners sit on the interface. Downstream stops scraping whatever is convenient.
- Named contracts on the models people actually consume.
- First certified exec path (often revenue or activation).
- On-call runbook with blast radius on one critical DAG.
One graph
Infra and metrics in the same place
The catalog AE uses is the same map on-call uses at 2am. Company Brain cites real nodes, not invented tables.
- Shared owners and blast radius.
- RBAC and audit on every answer.
- Human in the loop before any write path.
Hygiene
Kill unused paths on purpose
If nothing trusted reads it, schedule the sunset. The graph makes unused visible. That is the point.
- Unused extracts have nowhere to hide.
- Named sunset dates, not folklore.
- Less sprawl, lower bill, cleaner on-call.
Resources
IT & Data resources
Role playbooks plus two articles for the people who run the stack.
Playbooks
Data Platform Engineer
DAG debugging, connector health, and blast radius when a job retries.
Head of Data
Stack coverage, onboarding in days, and a revenue number you can show.
BI & Analytics Lead
Certified metrics in every tool. Reconciliation when finance flags a 2% delta.
ML Engineer
Feature provenance, drift alerts, and which models consume a column.
Data Governance Lead
Policy, ownership, and audit trails that stay current with the stack.