R&D experiment. It can change or be removed at any time and is not reviewed or approved. Do not rely on it for decisions. Read the full warning.
R&D / Banking AI Evaluation Library
Datasets and sources
Public, synthetic and restricted sources that could support an evaluation, with their limits.
Dodone links to the publisher and does not host or copy this data. Read the licence on the publisher page before you use it.
8 of 8 sources
BIAN semantic APIs (release 14.0.0)
OpenAPI and AsyncAPI files that describe the service operations of each BIAN service domain. They describe interfaces, not banking cases or answers.
BIAN Service Landscape v14.0
The full BIAN reference landscape, with business areas, business domains and service domains. It is a classification, not a set of evaluation tasks.
ISO 20022 message catalogue
Approved ISO 20022 message definitions and their versions, including payment messages. They define structure and rules, not bank outcomes.
BANKING77
Short customer service queries about banking, each labelled with one of 77 intents. It is made for fine-grained intent detection.
τ-bench banking_knowledge domain
A simulated customer service domain where an agent uses tools and searches policy documents, with tasks that have defined outcomes.
AMLworld (IBM synthetic AML transactions)
Synthetic transaction data sets in several sizes, where each transaction is labelled as laundering or not. The paper calls them IT-AML.
PaySim synthetic mobile money transactions
Synthetic transactions from the PaySim mobile money simulator, with a fraud flag on each transaction.
TAT-QA
Questions over tables and text from real financial reports, with answers and the numeric reasoning needed to reach them.