Skip to main content
KaireonAI ships with a curated dataset pack sourced from a well-known public dataset. The pack creates a complete, working decisioning environment — categories, offers, channels, creatives, schemas, models, qualification rules, contact policies, a decision flow, interaction history, and more — so you can start exploring immediately.

​
Available Dataset

DatasetSourceFocus
Starbucks OffersKaggle (CC0)Full Next-Best-Action (NBA) pipeline, multi-channel delivery, Thompson Bandit, segments
Only one dataset can be loaded at a time.

​
What Gets Created

The Starbucks dataset pack creates the full set of entities needed to run the decisioning pipeline end to end:
  • Categories and sub-categories — Top-level groupings (Acquisition, Retention, Engagement) with typed sub-categories beneath them.
  • Offers — 10 decision candidates with descriptions, priority scores, eligibility rules, budget caps, and tags. Each offer is assigned to a category and sub-category.
  • Channels — 6 delivery channels (web, email, mobile push, social, batch email, manual outreach) with channel type, delivery mode, and impression mode configured.
  • Creatives — 60 content variants (10 offers across 6 channels), each with channel-appropriate content (email subject lines, push deep links, web banners, social posts).
  • Schemas with fields — 3 entity schemas that create real PostgreSQL tables: a customer table (demographics, tenure, income), an offer portfolio table, and an events table. Each field has a data type, nullability, and display name.
  • Sample data rows — 100 synthetic customer rows, generated with deterministic random seeds so results are reproducible.
  • Qualification rules — 5 attribute-based conditions that filter candidates (e.g. minimum age, minimum income, loyalty membership).
  • Contact policies — 5 frequency caps and cooldown rules to prevent customer fatigue (e.g. max 3 offers per day, 24-hour cooldown between same-offer deliveries).
  • Algorithm models — 3 scoring models (scorecard, Bayesian, Thompson Bandit) trained on the dataset so you can compare approaches.
  • Experiments — A champion/challenger experiment splitting traffic between models.
  • Decision flow — A fully configured pipeline with inventory selection, qualification filtering, propensity scoring, ranking, and channel targeting.
  • Segments — Customer segments defined by filter conditions (e.g. high-income members with 365+ days tenure), materialized as SQL views with customer counts.
  • Interaction history — 500 synthetic interaction records (impressions, clicks, accepts, conversions, dismissals) with interaction summaries materialized for dashboard reporting.
  • Outcome types — Standard outcome taxonomy (impression, click, accept, convert, dismiss) with classifications.

​
How to Load

​
From the UI

  1. Navigate to Settings > Sample Data.
  2. The dataset appears as a card showing its name, description, testing focus tags, and entity counts.
  3. Click Load Dataset. If data is already loaded, the button reads Replace & Load and handles removal automatically.
  4. Wait for the confirmation toast. Loading takes a few seconds.

​
From the API

Send a POST request to load the Starbucks dataset: 
curl -X POST https://your-instance/api/v1/seed-dataset/starbucks \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "X-Requested-With: XMLHttpRequest"
To check which dataset is currently loaded: 
curl https://your-instance/api/v1/seed-dataset \
  -H "Authorization: Bearer YOUR_API_KEY"
The response includes entity counts and a currentlyLoaded field.
The seed endpoint requires the admin role and is rate-limited to 5 requests per minute.

​
Exploring the Data

Once the dataset is loaded, here are suggested first steps:
  1. Make a Recommend API call — Send a request to POST /api/v1/recommend with a customer ID from the loaded dataset (e.g. SBX-000042). The response shows which offers the pipeline selected, their scores, and personalized creative content.
  2. Browse Offers in Studio — Go to Studio > Offers to see all 10 offers that were created. Click into one to see its eligibility rules, budget configuration, and linked creatives.
  3. Open the Decision Flow — Navigate to Studio > Decision Flows to see the pre-built pipeline. It has inventory, qualification, scoring, ranking, and targeting stages already configured.
  4. Check the Operations dashboard — Go to Dashboards > Operations to see interaction history metrics, delivery counts, and outcome breakdowns populated from the synthetic interaction data.
  5. Explore algorithm models — Visit Algorithms > Models to see the 3 trained models. Compare the scorecard rules, Bayesian probabilities, and Thompson Bandit arm parameters.
  6. View a customer segment — Check segments under the segment builder to see the filter conditions and customer count.

​
Resetting

​
Remove the dataset

From the UI, click the Remove button on the loaded dataset card in Settings > Sample Data. Confirm the dialog. This deletes all entities created by the dataset, including schemas (and their PostgreSQL tables), interaction history, models, and experiments. From the API: 
curl -X DELETE https://your-instance/api/v1/seed-dataset/starbucks \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "X-Requested-With: XMLHttpRequest"

​
Full database reset

To drop all tables and start completely fresh: 
npm run db:reset
This drops all tables and recreates them. All data — not just sample data — will be lost.

Platform Walkthrough

Build a complete decisioning setup from scratch, step by step.

Quickstart

Get the platform running locally in under five minutes.

API Tutorial

Learn the Recommend and Respond APIs with hands-on examples.

Decisioning Studio

Explore the Studio components that the sample data created.