{"repo":"dynamiq-ai/pragmatiq","free":true,"listed":false,"github":"https://github.com/dynamiq-ai/pragmatiq","clone":"git clone https://github.com/dynamiq-ai/pragmatiq.git","description":null,"language":"Python","stars":41,"topics":["anti-money-laundering","banking","behavioral-modeling","churn-prediction","credit-scoring","deep-learning","embeddings","event-sequences","fintech","foundation-models"],"license":"Apache-2.0","category":"machine-learning","readme_excerpt":"pragmatiq pragmatiq is a developer-friendly implementation of ideas described in the PRAGMA paper (arXiv:2604.08649) by Ostroukhov et al. It turns user histories made of timestamped key-value events into embeddings that downstream teams can use for probes, LoRA fine-tuning, graph-based AML experiments, explainability, and serving. The repository includes the full stack: a deterministic synthetic banking data generator, tokenizer, padding-free PyTorch model, training pipeline, batch embedding, ONNX/Triton serving, notebooks, and a Streamlit demo. It is built to run on CPU first; CUDA and flash-attn are accelerations, not requirements. pragmatiq is an independent implementation inspired by the PRAGMA paper (arXiv 2604.08649). It is not affiliated with or endorsed by Revolut. Contents - Why this exists - Quickstart - What is included - Architecture - Repository map - Workflows - Using your own data - Training on GPU - Extending pragmatiq - AML over the transfer graph - Model sizes - PRAGMA+Nemotron text-embedding variant - Defaults where the paper is silent - Serving with Triton - Streamlit demo - Observability - Notebooks - Development - Citation - License Why this exists The PRAGMA paper (arXiv:2604.08649) describes a practical foundation-model recipe for heterogeneous banking behavior: transactions, app sessions, trades, communications, profiles, and transfer relationships. The paper does not ship reference code. pragmatiq provides an open implementation that researchers, ML ","default_branch":null,"files":null,"tree":[],"storefront":"/r/dynamiq-ai","claimed":false,"request_supported":{"post":"https://gitbuyer.com/r/dynamiq-ai/pragmatiq/request-supported","requests":0},"note":"indexed from public GitHub; nothing is for sale on this page. Clone it from GitHub. Paid listings live at /search."}