Fraud and anomaly detection — Banking
Graph and sequence models catching fraud and process anomalies in real time. Delivered for a banking client in Kutch.
- Location
- Kutch
- Year
- 2023
- Duration
- 4 weeks
- Team
- 14 engineers
- 85%
- fraud caught
- −66%
- false positives
- 20k
- tx per second
The challenge
The banking team was relying on manual effort that could not keep up with volume, and every miss had a direct cost.
What we built
We owned the whole stack — hardware, firmware, models and dashboards — using XGBoost, Graph ML, Kafka so there was one accountable team.
Delivery timeline
Phase 1
Discovery sprint and site survey
Phase 2
Prototype built and tested in the Motera lab
Phase 3
Pilot on site with before/after measurement
Phase 4
Scale-out, training and handover
Highlights
- Remote monitoring and over-the-air updates included
- Designed, built and supported from Motera, Ahmedabad
- Operator training delivered in Gujarati, Hindi and English
- Runs on-premises; no data leaves the site unless agreed
Related AI projects
AIDeliveredInsurance · GIFT City · 2022
Fraud and anomaly detection — Insurance
Graph and sequence models catching fraud and process anomalies in real time. Delivered for a insurance client in GIFT City.
- 73%
- fraud caught
- −46%
- false positives
- 12k
- tx per second
AIDeliveredE-commerce · Jamnagar · 2022
Fraud and anomaly detection — E-commerce
Graph and sequence models catching fraud and process anomalies in real time. Delivered for a e-commerce client in Jamnagar.
- 89%
- fraud caught
- −68%
- false positives
- 26k
- tx per second
AILiveTelecom · Kutch · 2023
Fraud and anomaly detection — Telecom
Graph and sequence models catching fraud and process anomalies in real time. Delivered for a telecom client in Kutch.
- 90%
- fraud caught
- −43%
- false positives
- 35k
- tx per second
AIPilotUtilities · Chennai · 2024
Fraud and anomaly detection — Utilities
Graph and sequence models catching fraud and process anomalies in real time. Delivered for a utilities client in Chennai.
- 81%
- fraud caught
- −48%
- false positives
- 16k
- tx per second
Have a problem that needs a machine built for it?
Tell us the constraint — line speed, accuracy, environment, deadline. An engineer will read it and reply with a straight answer, including when the answer is that you don't need us.
We reply to every enquiry within one business day.