User journeys

This page maps three end-to-end paths through Heimdall by RejiCo — each with exact sidebar clicks in the app at heimdallapp.org.

New account?

Create your account first. After sign-in you land on Dashboard (/dashboard) with an onboarding card if you have no data or models yet.

Which path should I take?

Your situation	Path	Time estimate
Production CSV or Excel → deployed API	Path A	~30–45 min
Just exploring — no file ready	Path B	~10 min
Messy data — profile before modeling	Path C	~45–60 min
Time series with dates	Path A, but publish Gold for Forecast → Forecast wizard	~30–45 min
Single text/image API call	Unstructured suite — not Lake	~5 min

Path A: CSV → Lake → Gold → ML → API

Best for

Your own spreadsheet or export, production use, and anything you will monitor in Usage.

Ingest to Bronze (~5 min)

1. Sidebar → Data Warehouse → Lake (or Dashboard → Lake catalog).
2. Click Add data (top right).
3. Choose Structured → upload .csv or .xlsx, enter a table name.
4. When upload completes, your table appears under the Bronze tab in the left catalog.
5. Click the table to preview rows and column types.

Detail: Add your first table

File size limit

Files over 100 MB are rejected. Split large exports or aggregate before upload.

Publish Gold (~5 min)

1. With the Bronze table open, click Publish Gold.
2. Name the artifact, select Machine learning, pick your target column.
3. Click Validate — fix any errors (needs 100+ rows and a valid target).
4. Publish. The artifact appears under Gold with status ready.

Detail: Publish for modeling

Train ML (~10–20 min)

1. Sidebar → Data Intelligence → Machine Learning.
2. Click New model (opens the training wizard).
3. Under Published dataset (Gold), select your Gold artifact.
4. Confirm target column → name the model → Train.
5. Watch the Training queue tab; open the model when status is complete.

Detail: Build an ML model

Deploy and call (~5 min)

1. Open your model at /ml/{id}.
2. Click Generate API key — copy it immediately (cannot be retrieved later).
3. Open the API Integration tab for endpoint URL and sample code.

Detail: ML deployment · API integration

Monitor

• Account trends: sidebar → Usage → Data Intelligence tab.
• Per-model log: model detail page → Usage section.

Detail: Monitoring & usage

---

Path B: Sample ML — skip Lake

Best for

First login, demos, and learning the wizard. Not for production.

Samples only

You cannot upload your own CSV directly in the ML wizard. Production data must go through Lake → Gold (Path A). Samples are the only way to skip Lake.

Steps (~10 min)

1. Sidebar → Data Intelligence → Machine Learning.
2. Click New model.
3. Scroll past the Gold picker → click Sample datasets.
4. Pick Housing (regression, target MEDV) or Iris (classification, target Species).
5. Confirm target → name model → Train.
6. When complete, deploy as in Path A step 4.

Detail: Quick start: sample

---

Path C: Lab profile → Silver → ML

Best for

Data quality unknown, joins, filters, or charts before modeling.

Bronze ingest

Same as Path A — get data into Bronze.

Explore in Lab (~15–30 min)

1. Sidebar → Data Warehouse → Lab (or Lake page → Open Lab).
2. Create a notebook → run list_tables() to see catalog names.
3. Load data: orders = lake("bronze", "Your Table Name").
4. Profile dtypes, nulls, outliers; chart with matplotlib/plotly.
5. When satisfied, Save result as Silver (creates a new Silver dataset — never overwrites Bronze).

Detail: Heimdall Lab

UI shortcut

The Lab Lake catalog sidebar inserts lake() snippets when you click a table name.

Silver → Gold → ML

1. Data → Silver tab → open your saved dataset.
2. Publish Gold (ML or Forecast) → validate → publish.
3. Machine Learning → New model → select Gold → train → deploy.

Optional: skip Lab if Bronze is already clean — promote to Silver in the Lake UI instead.

---

Lake vs direct ML upload

	Lake → Gold	Sample in ML wizard
Your CSV/Excel	✅ Required	❌ Not supported
Validation before train	✅ Row counts, target, dtypes	⚠️ Samples only
Reuse across ML & Forecast	✅ One Gold artifact	❌
Time to first API	~30–45 min	~10 min

---

ML vs Forecast vs Loop

Product	Sidebar	Use when	Data source today
Machine Learning	Data Intelligence → Machine Learning	Predict a category or number from features	Gold (ML) or samples
Forecast	Data Intelligence → Forecast	Predict future values over time	Gold (Forecast) or AirPassengers sample
Loop	Data Intelligence → Loop	Model must improve from every live prediction	Loop wizard (not Lake Gold yet)

Full comparison: What is Heimdall? — choose your product

---

Unstructured: Read/Vision vs Lake zip

Need	Path
One document or image, live API	Settings → API keys → Read/Vision
Hundreds of labeled images/text for training	Data → Add data → Unstructured → zip ingest

---

Next steps

• Quick start: your data — condensed Path A
• Quick start: sample — condensed Path B
• What is Heimdall? — product map & glossary
• Monitoring — after your first deploy