Why should you care?

Product teams want funnels and predictions on product behavior — not just static BI tables.

See it live — copy this example

Open mongosh or MongoDB Compass, select database nosqlverse, then run the example. Change one field and run again.

db.events.insertMany([
  { userId: "u1", type: "view", itemId: "p9", at: ISODate("2026-07-19T09:00:00Z") },
  { userId: "u1", type: "purchase", itemId: "p9", amount: 999, at: ISODate("2026-07-19T09:05:00Z") }
])
db.events.aggregate([
  { $match: { at: { $gte: ISODate("2026-07-19") } } },
  { $group: { _id: { userId: "$userId", type: "$type" }, n: { $sum: 1 }, spend: { $sum: { $ifNull: ["$amount", 0] } } } }
])
db.userFeatures.updateOne(
  { userId: "u1" },
  { $set: { views: 1, purchases: 1, spend: 999, embedding: [0.1, 0.2, 0.3], updatedAt: new Date() } },
  { upsert: true }
)

Run Example »

Edit the code below and click Run to see the result in Toolliyo’s live editor.

What happened?

• events is the append-only fact table.
• Aggregation builds counts/spend.
• userFeatures materializes a feature row (and optional embedding) for models and dashboards to read quickly.

Practice next

1. Insert mixed view/purchase events.
2. Run the group pipeline.
3. Upsert a feature document per user.
4. Compute funnel: views → purchases with $facet.
5. Index { userId: 1, at: -1 } on events.

Remember

Store events; materialize features. Aggregations power AI inputs. Optional vectors enable similarity.

Recommendation feature store

A retail app refreshes userFeatures nightly for “similar shoppers”.

Outcome: Models train on stable features without scanning raw clickstreams live.