Rasēd (Rased)

Rasēd (Rased) — Cross-institution fraud graph corpus: device and session fingerprints, account-to-identity-to-device relationship edges, transaction-flow subgraphs, and — most valuably — the accumulated adjudicated verdicts. Every investigator closure on a Mule Score alert or Ring Cluster is a human-confirmed ground-truth label for mule / not-mule and for collusive-ring membership, plus matched explainability rationales (natural-language justification paired to each alert). Seed-stage, ~20-person Riyadh-based vendor with a full fraud/compliance suite spanning six regulated verticals and a par…

Rasēd (Rased) offers (Alternative, Sentiment) — Derived from the integration surface rather than printed: per deployment the graph is built over full transaction data lakes plus session and device telemetry, so the corpus is transaction-scale (millions to hundreds of millions of edges at a mid-size retail bank) with a much smaller but far more valuable layer of adjudicated alert/ring verdicts on top — request further research for the aggregated cross-customer edge and alert counts.

Training and benchmarking graph neural networks for first-party fraud and AML, money-mule and ring-detection model development, synthetic-fraud-graph generation, investigator-triage automation, LLM fine-tuning for alert explainability and suspicious-activity-report drafting (from the rationale corpus), and cross-institution mule-typology benchmarking

The data is with 2 years of history.

Coverage spans Other, US, UK; Application Software, Diversified Banks; alternative, sentiment; equities.