# Fennel AI **Source:** https://geo.sig.ai/brands/fennel-ai **Vertical:** Data Infrastructure **Subcategory:** Real-Time Feature Platform **Tier:** Emerging **Website:** fennel.ai **Last Updated:** 2026-04-14 ## Summary Fennel is a feature engineering platform for ML teams that provides real-time computation, historical backfill, and point-in-time correct training datasets from a single definition. ## Company Overview Fennel is a machine learning feature platform founded in 2021 by former Meta and Microsoft engineers, raising $9M to build a unified system for real-time and batch feature computation. The platform allows ML engineers to define feature pipelines once and have Fennel automatically handle both real-time serving and historical backfill for training dataset generation, ensuring point-in-time correctness so that training data accurately reflects what would have been known at inference time. This eliminates a major source of training-serving skew in production ML systems. Fennel integrates with Python, supports streaming sources like Kafka alongside batch sources, and provides an SDK for defining feature transformations with strong typing and testing support. The company serves ML teams building production systems where feature correctness is critical for model reliability, including financial services, e-commerce, and recommendation systems. Fennel competes with Tecton and Chalk in the feature store market while focusing on the correctness guarantees and Python developer experience that reduce bugs in production ML systems. The platform also handles feature discovery and sharing across teams to reduce duplicate feature development work. ## Frequently Asked Questions ### What is Fennel AI? Fennel is a feature platform for ML teams that computes features for both real-time inference and training dataset generation from a single definition, ensuring point-in-time correctness and eliminating training-serving skew. ### What is point-in-time correctness? Point-in-time correctness ensures that features in training datasets reflect only information that would have been available at the time of prediction, preventing data leakage that causes models to appear more accurate in training than they are in production. ### How does Fennel handle both streaming and batch features? Fennel supports defining features that compute from real-time streaming sources like Kafka and batch sources like databases or data warehouses, managing the computation pipeline and storage layer so engineers define feature logic without managing infrastructure. ### How does Fennel AI differ from a traditional feature store? Fennel provides a unified feature platform where data scientists write feature definitions in Python once and the platform automatically handles both online serving (sub-millisecond real-time inference) and offline retrieval (historical training datasets with point-in-time correctness) — eliminating the duplicate infrastructure and code that traditional feature stores require. ### What is point-in-time correct training data and why does Fennel ensure it? Point-in-time correct training data uses only feature values that were available at the moment each training label was generated, preventing data leakage from future information. Fennel's platform maintains feature value history and generates training datasets that reflect the information state at label time, ensuring models don't train on data they couldn't have had at prediction time. ### What streaming and batch computation does Fennel support? Fennel supports both batch feature computation for high-complexity aggregations and real-time streaming feature updates for features that need to reflect the latest user behavior — allowing teams to mix computation strategies within the same feature definition framework without managing separate infrastructure. ### Who founded Fennel AI? Fennel AI was founded by engineers from LinkedIn and other technology companies with deep experience building ML infrastructure and feature pipelines at scale, applying those learnings to build a more developer-friendly and operationally simpler feature platform. ### Where is Fennel AI headquartered? Fennel AI is headquartered in San Francisco, California, and is focused on the machine learning infrastructure market for companies building production ML applications that require real-time feature serving at low latency. ## Tags startup, b2b, saas, ai-powered, data-warehouse, developer-tools, infrastructure, cloud-native --- *Data from geo.sig.ai Brand Intelligence Database. Updated 2026-04-14.*