Quickstart
Go from an empty project to a served, auditable prediction — in eight steps, entirely through the public API. You will upload a CSV of e-commerce events, approve what the model may see, define a churn-adjacent label, train a behavioral model on those events, and read back a per-user embedding and score.
Every code block in this quickstart is a real file, executed against a
seeded test project in CI on every change — the samples cannot drift from
the platform. Each step shows the curl form to run and a form for the planned
Python SDK.
What you need
- A Tessera account, project, and a
tsk_test_…API key — create them first. - The sample dataset:
tessera-sample-v1.zip — three
event domains (e-commerce orders, media streaming, logistics)
plus a label file, deterministically generated; its
manifest.jsoncarries per-file checksums and the dataset fingerprint. curlandpython3.
The samples read four environment variables — set them once:
TESSERA_API_BASE # https://api.staging.tessera.zone/v1 (or your stack)
TESSERA_INGEST_BASE # https://ingest.staging.tessera.zone/v1
TESSERA_API_KEY # tsk_test_…
TESSERA_PROJECT_ID # proj_…
The eight steps
| Step | What happens | Concept it teaches |
|---|---|---|
| 0. Create a project | org → project → API key | tenancy |
| 1. Send your first events | two events through the ingestion endpoint | entities & events |
| 2. Upload historical data | the e-commerce CSV → canonical events | column mapping |
| 3. Approve the contract | review what the model may see | schema contracts |
| 4. Define a label | repeat_purchase_30d from the sample label file | labels & masking |
| 5. Build the dataset | leakage-gated, point-in-time training data | leakage gates |
| 6. Train | a real training run + the six promotion gates | promotion policy |
| 7. Activate & batch score | bundle → slot → refresh | fingerprints |
| 8. Read the state | embedding + score + freshness per user | serving |
Prefer to see where you're headed first? Jump to step 8's output — one JSON object with an embedding, a score keyed by your label, and the fingerprint trail that proves where both came from.