Generates massive synthetic financial transaction graphs with cyclic fraud patterns for benchmarking graph ML models.
Build benchmark-ready financial transaction graphs with embedded multi-hop money laundering patterns.
gen-fraud-graph is a Python tool that generates synthetic financial transaction datasets with built-in money-laundering rings. You can use it to benchmark graph neural networks, evaluate anti-money laundering algorithms, and load-test graph databases without exposing sensitive financial records.
The tool creates synthetic account nodes, normal transaction edges, and cyclic fraud rings spanning four to seven hops. It exports datasets in generic CSV format or AWS Neptune bulk-load syntax, complete with optional vector embeddings from local or OpenAI models. Multi-process parallel generation allows you to scale outputs from one thousand to over ten million accounts.
Data scientists, machine learning engineers, and database architects working on financial crime detection will benefit most from this tool. It is not intended for end users seeking pre-trained models or a graphical user interface, as configuration and execution take place entirely in the command line or via Python scripts.
Generates multi-hop cyclic money-laundering patterns with configurable depth from 4 to 7 transaction hops.
Supports random vectors, local SentenceTransformers models, or OpenAI API integrations for node and edge features.
Distributes generator execution across multi-process workers to create datasets up to tens of millions of records.
Exports generated CSV files formatted directly for AWS Neptune bulk-loading alongside standard generic CSV output.
gen-fraud-graph is free to run under the Apache-2.0 licence. Your only costs stem from the computing infrastructure and storage required for large generations, along with optional OpenAI API costs if you select external cloud embeddings.
880 open remote AI-training roles, updated November 15, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
These reviewed AI tools do the same job instantly — from our scored directory.
Output size ranges from approximately 2 MB at 1,000 accounts (scale 0.0001) up to 20 GB or more for 10 million accounts (scale 1.0).
No. You can use the default 'fake' provider for fast random vectors or the 'local' option to generate embeddings using local SentenceTransformers models without API fees.
See Zekai first in Google