Real-time features, AI search, Agentic similarities

2026-02-02 · Source: MLOps.community · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Cloud Computing & IT Infrastructure · Depth: Advanced, extended

Summary

Kronon, originally developed at Airbnb as Zipline, is a feature platform designed to address the complex data engineering challenges of real-time feature computation for AI and ML applications. Unlike traditional feature stores that offload compute, Kronon focuses on handling streaming aggregations, online serving, and offline training with consistency, providing an easy API for users. Its genesis was driven by the need to combat payments fraud at Airbnb, requiring rapid feature engineering and testing. The platform was later open-sourced in collaboration with Stripe, another payments-heavy company, and subsequently found applications in search personalization and product metrics (e.g., real-time star ratings). The founders observed that existing solutions like Feast and Feather failed to fully address the "compute" problem, particularly streaming and window aggregations, which Kronon claims to have uniquely solved. The company is now expanding its focus upstream to foundational data infrastructure, including data governance, privacy, and making open-source technologies like Iceberg easier to adopt, while also integrating native embedding support for LLM-driven use cases like customer support and virtual travel agents.

Key takeaway

For AI Architects building real-time ML systems, Kronon offers a robust solution to the persistent challenges of feature computation and consistency across online inference and offline training. You should evaluate Kronon's open-source platform for its ability to simplify complex streaming aggregations and point-in-time correct data generation, especially if your current stack involves duct-taping multiple big data technologies. Consider its expanding capabilities in data governance and native embedding support for LLM-driven applications to streamline your data infrastructure and reduce technical debt.

Key insights

Kronon uniquely solves real-time feature computation challenges for AI/ML by integrating streaming, online serving, and offline training.

Principles

• Compute is the hardest part of data engineering for AI/ML.
• Point-in-time correct training data generation is an n-cube problem.
• Battle-testing at scale across multiple companies is crucial for robust solutions.

Method

Kronon takes raw data, produces features, and serves them online for inference and offline for training, abstracting complex underlying technologies like Spark, Flink, and BigTable via a SQL-like API.

In practice

• Use Kronon for real-time fraud detection and search personalization.
• Integrate Kronon for consistent online/offline feature serving.
• Explore Kronon's native embedding support for LLM context engineering.

Topics

• Feature Platforms
• Real-time Feature Engineering
• Streaming Aggregation
• AI/LLM Applications
• Data Infrastructure

Best for: AI Architect, Machine Learning Engineer, Data Engineer, MLOps Engineer

Related on AIssential

• The Hardware Bottleneck AI Can’t Fix — Hardware engineering's lack of software-like tooling creates a bottleneck, hindering AI's impact and requiring specialized data platforms.
• Feature Stores from Scratch: A Minimal Working Implementation — A feature store's five core components provide consistent, low-latency feature access for both traditional ML and LLM applications.
• MongoDB - Mistral AI — Mistral AI's SMoE architecture combined with MongoDB's unified data platform enables efficient, real-time enterprise AI applications.
• Building Real-Time Product Search on Databricks — Modern product search requires an integrated platform for real-time ingestion, retrieval, and refinement, balancing speed with relevance and business outcomes.
• Is there still a role for a feature store — Existing feature stores often fall short in addressing complex compute challenges for AI/ML data engineering.
• The emergence of the web data infrastructure layer for AI — Specialized web data infrastructure is critical for AI to access real-time, trustworthy, and contextually relevant information at scale.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by MLOps.community.