Financial & AI Glossary
Modern quantitative finance sits at the intersection of three worlds: traditional finance, machine learning, and data engineering. Each speaks its own language. A quant may never have written an embedding; an ML engineer may not know an accrual from a regime. This glossary closes that gap.
Every entry is plain-English first, then grounded in how VectorFin actually surfaces the concept: which API endpoint returns it, which Iceberg column it lands in, how it behaves under point-in-time queries. These are not abstract textbook definitions; they describe the data you get when you pull from the platform. Pick a term from the list on the left.
What it covers
ML & AI
Embeddings, vector similarity, dense retrieval, and Retrieval-Augmented Generation: the machinery that turns unstructured filings and earnings calls into model-ready vectors and grounds LLM answers in citable source text.
Quant finance
The accounting-based scores VectorFin computes weekly: the Piotroski F-Score, Altman Z-Score, Beneish M-Score, and Sloan accrual, plus HMM market-regime detection. Each maps to a typed column in the signals table.
Data engineering
How the data is stored and served: Apache Iceberg tables on GCS, the Apache Polaris REST catalog that vends credentials to BigQuery and Snowflake, and the bitemporal model (effective and knowledge timestamps) that makes point-in-time queries safe for backtests.
Finance & NLP
The source material itself: earnings-call transcripts, fiscal periods, and the semantic-search patterns used to query embeddings of them.
How to read an entry
Each term opens with a one-line definition, then a plain-English explanation, a technical definition where it helps, and a “How VectorFin uses this” section tying the concept to real tables, columns, and endpoints. Related terms link across domains, so you can follow a thread from an embedding to cosine similarity to the RAG pipeline that consumes both.