User journeys

Pick one path. The product pages linked from each step are the detailed version. Do not follow a second quick start for the same task.

Situation	Path
Your CSV or Excel, then an API	Path A
A time series	Path A, Forecast
No file yet	Path B
Profile or join before training	Path C
One text or image call	Read and Vision

A new account lands on Dashboard (/dashboard). Create an account if you do not have one.

Path A: your file

Files over 500 MB are rejected. Split the export first.

Bronze

1. Data Warehouse → Lake (/data). Dashboard → Lake catalog opens the same page.
2. Add data → Structured.
3. Upload .csv or .xlsx and enter a table name.
4. Open the table under the Bronze tab.

Add your first table

Gold

1. With the table open, click Publish gold.
2. Name the artifact. Select Machine learning and the target column. You need 100+ rows.
3. Validate, then Publish. The artifact is under Gold with status ready.

A clean file can go from bronze to gold. Use silver or Lab when you need a join, a filter, or a profile first.

Publish for modeling

Train and call

1. Data Intelligence → Machine Learning → New model.
2. Published dataset (Gold) → select the artifact.
3. Confirm the target column, name the model, Train.
4. Open it from Training queue at /ml/{id}.
5. Generate API key. Copy it on that screen. It is not shown again. A new key replaces the old one.
6. API Integration has the request.

Build a model · Deploy

Account charts are Usage → Data Intelligence. The model page has its own Usage section.

Forecast

Publish gold with the Forecast use case instead of Machine learning. You need 30+ rows. Set Time column (forecast), or leave auto-detect. Then Data Intelligence → Forecast → New Forecast, pick that gold artifact, and set the interval to Hourly, Daily, Weekly, Monthly, or Quarterly.

Build a forecast

Path B: a sample

The ML wizard does not accept your CSV. Samples are the only way to train without Lake. Do not ship a sample model.

1. Data Intelligence → Machine Learning → New model.
2. Sample datasets.
3. Housing (regression, target MEDV) or Iris (classification, target Species).
4. Confirm the target, name the model, Train.
5. Deploy with the same key steps as Path A.

The forecast sample is AirPassengers, interval Monthly. Build a forecast.

Path C: Lab

1. Land the file in bronze, as in Path A.
2. Data Warehouse → Lab. Create a notebook. list_tables() prints the catalog. lake("bronze", "Your Table Name") loads a table. Clicking a name in the Lake catalog sidebar inserts that line.
3. Save result as silver. That creates a new silver dataset. It does not overwrite bronze. A cell can load up to 6 tables.
4. Open the silver table in Lake and publish gold, then train as in Path A.

Lab

Which product

Product	Use when
ML	A category or a number from columns
Forecast	Future values of a series
Loop	Each prediction later gets an outcome you send back
Resolve	A choice or a score from text you define, with no table
Read / Vision	One text or image, from Account → API keys

A labeled zip of images or .txt files is Lake → Add data → Unstructured, not the Read key. Zip layout.

What is Heimdall?