Act 1 · under the hood

BIW's data lives in 18+ places, and none of them agree

Real-time streams, change-data-capture, nightly batch, dropped files, and an existing Databricks lakehouse. A single warehouse load can't touch this. An open lakehouse can.

18 sources

across the sales org

4 velocities

stream · CDC · batch · file

1 open lakehouse

watsonx.data (Iceberg + Trino)

0 migrations

Databricks stays put

Act 2 · under the hood

One event, live — then joined with data we never moved

Inject a source event and watch it stream through Kafka, dual-write to the live store and the lakehouse, and get joined — zero-copy — with a table that stays inside Databricks. Then start the live stream to see Salesforce CDC and Slack events landing in real time.

Inject an event:

Source

Slack / Salesforce

→

Kafka

IBM Confluent

3 topics

→

Dual-write

watsonx.data

consumer

→

Cassandra

IBM DataStax

live "now"

Iceberg

watsonx.data

history "always"

⋈ Trino federated join — live + history + Databricks ◆ Databricks — read in place · 0 rows copied

SELECT s.seller_id, live.event_type, live.event_ts, hist.event_ts AS history_ts, db.event_ts AS databricks_ts
FROM cassandra.biw_live.biw_live_events live
FULL OUTER JOIN iceberg.biw.live_events hist ON live.seller_id = hist.seller_id
FULL OUTER JOIN databricks.main.biw_demo.seller_history db ON live.seller_id = db.seller_id -- in place, 0 rows copied

Live Kafka stream · Confluent

Salesforce CDC and Slack — arriving now

Watch real events land in two Confluent topics in real time. Every row below is a message that just hit Kafka.

Stream 30 seconds of live events into Confluent Kafka

Salesforce CDC · Slack · sellers: Rachel, Marcus, Maya

biw.salesforce.cdc 0

Salesforce CDC

biw.slack.events 0

Slack / collaboration

R

Rachel 0

salesforce

0

slack

0

M

Marcus 0

salesforce

0

slack

0

M

Maya 0

salesforce

0

slack

0

Live event feed waiting…

Click Start 30-second stream to begin

Events from Salesforce CDC · Slack will appear here as they land in Kafka

not started

cluster: pkc-zgp5j7.us-south1.gcp.confluent.cloud · SASL_SSL

Act 3 · through the glass

The seller sees one thing: a nudge that finally fits

One nudge, one action list — priced in points, justified by the signals that produced it, and delivered to the surface the seller already works in. Pick a seller, then pick where it lands.

🔒 Seller-facing — the platform stack (watsonx.data · Kafka · Trino) never appears on these surfaces

Where the nudge lands

Loading nudge…

ⓘ All seller data is 100% synthetic (SEED=20260812). Numbers shown are simulated for demonstration purposes. The underlying pipeline (Kafka → Cassandra + Iceberg → Trino federation) processes real engines.