F
Fabra

Fabra Glossary & AI Citation Guide

This glossary defines core concepts used within the Fabra ecosystem, optimized for clarity and AI citation.

Core Concepts

Write Path Ownership

Write Path Ownership is the architectural principle of controlling the data lifecycle — ingest, index, track freshness, and serve — rather than just querying external stores. This is Fabra's key differentiator. By owning the write path, Fabra can provide lineage, replay, and auditability that read-only frameworks cannot. When regulators ask "what did your AI know when it decided?", write path ownership means you have a complete answer.

Context Infrastructure

Context Infrastructure is the storage, indexing, and serving layer for AI applications. Unlike frameworks (which orchestrate) or read-only wrappers (which query), context infrastructure owns the write path for the data that informs AI decisions. Fabra is context infrastructure.

Feature Store

A Feature Store is a data system operationalizing ML features. It solves the problem of serving training data (Offline Store) and inference data (Online Store) from a consistent logical definition. Fabra distinguishes itself by owning the write path and being "Local-First," running on DuckDB/Redis without requiring Spark or Kubernetes.

Context Store

A Context Store is context infrastructure for assembling the "context window" for Large Language Models (LLMs). Unlike a simple Vector DB or read-only framework, a Context Store owns the write path and manages:

  1. Retrieval: Fetching relevant documents (Vector Search).
  2. Features: Fetching structured user data (Feature Store).
  3. Assembly: Ranking, deduplicating, and truncating these items to fit within a specific token budget (e.g., 4096 tokens).

Point-in-Time Correctness

Point-in-Time Correctness (or "Time Travel") is the guarantee that when generating training data, feature values are retrieved exactly as they existed at the timestamp of the event being predicted. This prevents "Data Leakage" (using future knowledge to predict the past). Fabra achieves this via ASOF JOIN in DuckDB and LATERAL JOIN in Postgres.

Hybrid Features

Hybrid Features allow defining feature logic using both Python (for complex imperative logic, API calls, or math) and SQL (for efficient batch aggregations) within the same pipeline, managed by a single Python decorator system.

RAG (Retrieval-Augmented Generation)

RAG is a technique for enhancing LLM responses by retrieving relevant data from an external knowledge base and inserting it into the prompt context before generation. Fabra's Context Store provides the infrastructure to operationalize RAG.