Real-Time Fraud
Intelligence Platform
Portfolio-grade fraud detection platform with graph analytics, ML training pipelines, and a local-first RAG assistant.
32
Tests (Local Run)
12
Local Services
2
Model Families
3
Primary Stores
Built for Proof, Not Hype
Every section maps to code in this repo. Heavy services run locally; the hosted demo is intentionally lightweight.
Graph Analytics
Neo4j-backed network analysis with PageRank and community detection to surface suspicious rings.
ML Training + Explainability
XGBoost and GAT training scripts with MLflow logs and SHAP explanations surfaced in the UI.
Streaming + Features
Kafka + Flink CEP for real-time patterns and Spark/Feast for feature engineering (local stack).
RAG Assistant
Local-first investigator copilot with Groq/Ollama options and pgvector-backed retrieval.
Feature Store
Feast feature repo with Redis cache and Postgres + pgvector for retrieval data.
Full-Stack
Next.js 16 + FastAPI + GraphQL with Docker Compose for local runs and free-tier demo targets.
Operational Snapshot
Local EvidenceEach item maps to code in this repo. Heavy services run locally; hosted demos are intentionally lightweight.
Fraud scoring and graph insights on demand.
The demo UI is backed by a FastAPI + GraphQL layer. Use REST for scoring and GraphQL for the fraud queue, explanations, and network data.
- REST scoring endpoint for transaction risk
- GraphQL fraud queue for analyst workflows
- SHAP explanations for model transparency
curl -X POST http://localhost:8000/score \
-H "Content-Type: application/json" \
-d '{
"transaction_id": "txn_123",
"amount": 740.25,
"merchant": "Retail"
}'{
"fraud_score": 0.87,
"decision": "review"
}