piyush314159/gnn-fraudnet

How piyush314159/gnn-fraudnet works

GNN-FraudNet serves transaction-fraud predictions and analyst investigations through a FastAPI API. An analyst or client submits a transaction node; the investigation agent coordinates prediction, explanation and graph analysis, then optionally retrieves fraud knowledge and generates an LLM report. The repository also includes dataset preparation, model training and evaluation capabilities. The workflow and some subsystem links are documented in the README; implementation excerpts are incomplete, so unverified links are dashed and uncited.

Main entry points: investigation_agent.py Investigation agent · report_generator.py LLM report generation · explainability.py Model explanations · graphsage.py GNN and baselines

Components 19

API and workflow

FastAPI endpoints api/main.py
Investigation agent src/agents/investigation_agent.py
Runtime configuration src/config.py

Data and models

Transaction graph data src/data_loader.py
Graph construction src/graph_builder.py
GNN and baselines src/models/graphsage.py

Investigation services

Fraud prediction src/services/prediction.py
Model explanations src/services/explainability.py
SHAP utilities src/explain.py
Neighborhood analysis src/services/graph_analysis.py
Investigation context src/services/investigation.py

Knowledge and reports

Fraud knowledge base src/rag/knowledge_base.py
Semantic retrieval src/rag/retriever.py
LLM report generation src/llm/report_generator.py
Report prompts src/llm/prompts.py

Training and evaluation

Model training src/train.py
Model evaluation src/evaluate.py

Other components

Analyst / client
LLM provider

Connections 21

Analyst / client → FastAPI endpoints: submits request api/main.py
FastAPI endpoints → Investigation agent: routes investigation api/main.py
Investigation agent → Fraud prediction: predicts node src/agents/investigation_agent.py
Investigation agent → Model explanations: explains prediction src/agents/investigation_agent.py
Investigation agent → Neighborhood analysis: analyzes neighborhood src/agents/investigation_agent.py
Investigation agent → Semantic retrieval: retrieves knowledge src/agents/investigation_agent.py
Investigation agent → LLM report generation: generates report src/agents/investigation_agent.py
Fraud prediction → GNN and baselines: runs model
Model explanations → SHAP utilities: runs SHAP src/services/explainability.py
Model explanations → GNN and baselines: explains model src/services/explainability.py
Semantic retrieval → Fraud knowledge base: queries collection src/rag/retriever.py
LLM report generation → Report prompts: formats prompt src/llm/report_generator.py
LLM report generation → LLM provider: requests completion src/llm/report_generator.py
LLM report generation → Semantic retrieval: uses retrieved context src/llm/report_generator.py
Investigation context → Fraud prediction: requests prediction src/services/investigation.py
Investigation context → Model explanations: requests explanation src/services/investigation.py
Investigation context → Neighborhood analysis: requests analysis src/services/investigation.py
Graph construction → Transaction graph data: prepares graph
Model training → Transaction graph data: trains on graph
Model training → GNN and baselines: trains models
Model evaluation → GNN and baselines: scores models

Drawn by GitDiagram on November 14, 2026 from the repository's file tree, README and source files. Read as Markdown ·