A real-time feature store is the difference between batch ML and ML that participates in user experience. Here is a working reference architecture.
- 01Define features as code (Feast feature views)
Source of truth is git, not ad-hoc SQL.
- 02Stream features into Redis from Kafka
Sub-100ms reads with simple infra.
- 03Materialize batch features nightly into the same Redis
Online and offline serve from the same store.
- 04Validate read parity between online and offline
Daily Great Expectations test suite.
- 05Wire feature serving into the inference path
Model gets features by entity ID; never reaches into source systems.
Feast Feature View
from feast import FeatureView, Entity, Field, FileSource
from feast.types import Float32, Int64
from datetime import timedelta
customer = Entity(name="customer_id", join_keys=["customer_id"])
customer_30d_features = FeatureView(
name="customer_30d_features",
entities=[customer],
ttl=timedelta(days=2),
schema=[
Field(name="orders_30d", dtype=Int64),
Field(name="aov_30d", dtype=Float32),
Field(name="returns_rate_30d", dtype=Float32),
Field(name="time_since_last_d", dtype=Float32),
],
online=True,
source=FileSource(
path="s3://features/customer_30d.parquet",
timestamp_field="event_ts",
),
)Ready to optimize your cloud or AI footprint?
Book a free 30-minute architecture review. We will deliver a written cost-and-architecture audit within 48 hours.
Need help with real-time feature store?
Ohveda runs free 30-minute architecture reviews. We will identify your top opportunities in writing within 48 hours — at no cost.