Heimdall Lake
Heimdall Lake is the managed medallion data catalog on Heimdall by RejiCo — where you ingest raw data (bronze), curate it (silver), and publish modeling-ready gold datasets before ML or Forecast training.
Before you start
- Sidebar → Data Warehouse → Lake (
/data) - Production ML/Forecast requires Gold — Bronze alone is not enough
Product page
Medallion layers
| Stage | Tab in /data | Meaning |
|---|---|---|
| Bronze | Bronze | Raw upload — CSV, Excel, zip job output |
| Silver | Silver | Joins, filters, Lab outputs |
| Gold | Gold | Validated for ML or Forecast training |
Add data (UI)
Lake page → Add data (top right):
| Option | Use for |
|---|---|
| Structured | CSV / Excel → Bronze |
| Unstructured | Labeled zip → Bronze table via ingest job |
| Create Silver dataset | Join/filter Bronze without Lab |
| Database | Coming soon — export CSV today |
Common paths
| Goal | Flow |
|---|---|
| CSV → ML API | Bronze → Gold (ML) → Machine Learning wizard → Path A |
| Profile first | Bronze → Lab → Silver → Gold → ML Path C |
| Time series | Bronze → Gold (Forecast) → Forecast wizard |
| Image/text corpus | Unstructured zip → Bronze → optional Gold → ML |
Lake vs ML wizard upload
| Lake → Gold | ML samples only | |
|---|---|---|
| Your CSV | ✅ Required | ❌ |
| Validation | ✅ Before training | Samples only |
| Reuse in Forecast | ✅ | ❌ |
Direct CSV upload in the ML wizard is disabled (verified in app API).
FAQ
Unstructured zip vs Read/Vision APIs?
Zip = bulk training data in Lake. Read/Vision = single-request APIs. See Unstructured suite.
Legacy Forge URLs?
/forge redirects to /data?ingest=unstructured. Unstructured ingest is part of Lake.
Do not skip Gold for production
Training without Gold bypasses row-count and schema checks — failures happen mid-train.
Next steps
- User journeys
- Add your first table
- Publish for modeling
- Lab
- Sample quick start — skip Lake for demos