Course Objective & Scope
When data scientists write feature logic in SQL or Pandas for training and backend engineers rewrite that logic in Java or Go for production, the result is the most insidious failure mode in MLOps: training-serving skew. Worse still, naive historical joins easily leak future state into training rows, making models look stellar offline and fail in production.
This in-depth programme covers dual-storage (Offline + Online) Feature Store architectures, hands-on implementation with open-source Feast (alongside architectural comparisons with Tecton, Databricks and SageMaker Feature Store), point-in-time correct historical joins (ASOF joins), low-latency online serving via Redis/DynamoDB and streaming feature pipelines over Kafka. Participants build a shared feature registry and Feast reference pipeline for their organisation.
Who Should Attend?
- MLOps and data engineers building infrastructure for real-time recommendation, credit scoring, dynamic pricing and fraud models
- Data scientists standardising feature reuse and eliminating training-serving discrepancies across teams
- Platform architects bridging lakehouse/data-warehouse analytics with low-latency inference services
- Prerequisite: Python, strong SQL (window functions) and core data engineering concepts
Key Takeaways
- Eliminate training-serving skew by driving both historical training sets and live inference from a single feature definition
- Prevent future data leakage mathematically using point-in-time correct (ASOF) historical joins
- Deploy Feast with an Offline Store (Parquet/Postgres/BigQuery/Snowflake) and materialise latest values to an Online Store (Redis)
- Combine batch features, streaming aggregations (Kafka) and on-demand request-time transforms in one feature service
- Monitor feature freshness and serving latency SLAs while governing feature ownership and reuse
Curriculum
Training-Serving Skew, Time Leakage and Feature Store Anatomy
- Why neither a data warehouse nor a standalone Redis cache solves the dual-access problem on its own
- Offline Store, Online Store, Feature Registry, Materialisation Engine and Point-in-Time (ASOF) Join mechanics
Declarative Feature Definitions and Point-in-Time Training Sets with Feast
- Defining Entities, FeatureViews, Fields, TTLs and data sources declaratively in Python
- Generating leakage-free historical training datasets aligned to event_timestamp via get_historical_features
Online Store Materialisation (Redis) and Low-Latency Serving
- Syncing latest feature states into Redis/DynamoDB with feast materialize-incremental
- Fetching online feature vectors in single-digit milliseconds via get_online_features inside inference services
Combining Batch, Streaming and On-Demand Feature Pipelines
- Unifying daily dbt/Spark batch tables with 5-minute sliding-window aggregations pushed from Kafka/Flink
- Computing request-time contextual ratios and transforms using On-Demand Feature Views
Production Scaling, Freshness SLAs and Platform Selection
- Sizing Redis memory footprints, tuning TTLs, handling missing-entity defaults and alerting on stale features
- Architectural comparison matrix: open-source Feast versus Tecton, Databricks and SageMaker Feature Store
Other Courses in This Category
-
Pre-deployment fairness (Fairlearn), explainability (SHAP), probability calibration, uncertainty quantification (Conformal Prediction), guardrails, progressive rollouts and kill-switches, for ML and risk teams.
- Live Course
- 5 modules
-
Designing an enterprise AI platform unifying classical ML and GenAI/LLMs: compute/GPU orchestration, data and model layers, AI Gateways, multi-tenant isolation and reference architectures, for architects and platform leads.
- Live Course
- 5 modules
-
Designing an Internal Developer Platform (IDP) for data scientists and AI engineers: Backstage templates, on-demand notebook/GPU workspaces, Golden Path pipelines and self-service deployment, for platform teams.
- Live Course
- 5 modules
-
IAM/RBAC/ABAC for AI workloads, secret management with HashiCorp Vault and External Secrets, Policy-as-Code with OPA/Kyverno, supply-chain security (pickle/images) and network isolation, for platform and security teams.
- Live Course
- 5 modules
-
GPU and LLM token unit economics, cost visibility with DCGM/Kubecost/OpenCost, quota/priority scheduling with Kueue, MIG/fractional GPU sharing, Spot training and model routing, for AI platform and FinOps leads.
- Live Course
- 5 modules
-
Moving from Jupyter notebooks to production: modular Python code, DVC, MLflow, FastAPI packaging, Docker and GitHub Actions, for data scientists and engineers building a working end-to-end machine learning pipeline.
- Live Course
- 5 modules
Get in touch
