Vol. II · NPCI Hackathon 2026Blue Team · UPI Fraud Detection

Real-time fraud protection for India’s 640M+ daily UPI transactions

Varaksha.

UPI Fraud Defense Network

Rust · Random Forest · IsolationForest · NetworkX · edge-tts
Why This Matters

The Problem

1,750Cr

Lost to UPI fraud last year

FY 2024

— RBI Annual Report 2024

The Scale

640M

UPI transactions every single day

Real payments, real people

— NPCI

Model Performance

96.52%

Fraud Detection Accuracy (V2)

111K rows · RF + LightGBM fusion · ROC-AUC 0.9952

— Varaksha V2 · validated Mar 2026

Our Speed

<10ms

To deliver a fraud verdict

Faster than a human blink

— Varaksha Gateway · Criterion.rs

Who We Protect

44%

Indians can’t read an English fraud alert

Varaksha alerts in your language

— India BRICS Language Survey

Gateway Memory View
Rust DashMap Live Cam~320 TPS simulated · presentation mode

Hit

0

Miss

0

Intercept

0

Total

0

How We Designed This

Instant payments. Intelligent fraud defence.

The Challenge

Every UPI transaction settles in under 15 seconds. If your fraud model sits on that critical path — between the user and their bank — one slow inference call can stall payments for millions of people. That’s the tradeoff every legacy system forces you to accept.

How We Solve It

Our Rust gateway reads from a shared risk cache and returns ALLOW / FLAG / BLOCK in under 10 ms — without ever calling the ML model in real time. The Random Forest and graph agent run entirely in the background, writing better signals into the cache after every transaction.

Five Layers — One Verdict in Under 10msasync · off-path · consortium

L1

ML Engine

RF-300 · IsolationForest · SMOTE

Active

L2

Rust Gateway

DashMap · SHA-256 · <10ms

Active

L3

Graph Agent

NetworkX · Fan-out · Cycles

Active

L4

Alert Agent

LLM · Hindi alert · edge-tts

Active

L5

Dashboard

Real-time Verdict Stream

Active