Fraud Analytics Expert
LenDenClubMumbai, IndiaOn-Site
Needs a free Superjobs account. You'll finish on LenDenClub's site.
Skill score
Summary
This person builds and owns graph-based fraud ring detection models and face match and liveness models for identity verification, taking them from data through production inference. It suits a machine learning engineer with shipped fraud or identity models, hands-on GNN and computer vision experience, and comfort explaining alerts to investigators and risk leaders.
- Function
- Data Science
- Estimated salary
- Est. ₹9.5–14 LPA
- Estimated for a mid-level role, ~3 yrs experience, India. Not published by the employer.
- Work mode and location
- On-Site, Mumbai, India
- Experience
- 3+ yrs
Job description
Overview
You will build the AI that stops fraud at LenDenClub before money moves: graph models that expose fraud rings, and face match and liveness models that confirm a real, present person is behind every application. You own these models from data to production.
- Fraud rings: model borrowers, devices, phones, bank accounts, UPI handles, addresses and IPs as one graph, and catch coordinated networks that single-application rules miss.
- Identity: verify that the selfie matches the KYC photo and that the face is live, not a print, replay, mask, deepfake or injected camera feed.
- Real stakes: your models sit in the live onboarding and disbursal flow of an RBI-regulated lender, where a miss means a loss and a false alarm means a lost genuine customer.
Roles and responsibilities
- Design the entity graph: nodes, edges and features across applications, devices, bank accounts, UPI handles, contacts and addresses.
- Build GNN models (GraphSAGE, GAT, heterogeneous and temporal GNNs) for node classification, link prediction and fraud-ring detection.
- Combine GNNs with community detection (Louvain, Leiden), graph features and tabular models where they perform better.
- Build and tune face match on KYC photos (Aadhaar, PAN, DigiLocker, Video KYC captures) using ArcFace-class embeddings.
- Build passive and active liveness and presentation-attack detection against prints, replays, masks, deepfakes and virtual-camera injection.
- Measure performance by device tier, lighting and skin tone, and close gaps before they reach customers.
- Deploy low-latency inference for real-time onboarding, including on-device models (ONNX, TFLite) where needed.
- Work with risk, operations and investigation teams to label cases, explain alerts and turn confirmed fraud into training data.
Skills required
Top skills
- Machine learning
- Graph neural networks
- PyTorch Geometric
- Computer vision
All skills (10)
- Face recognition
- Face anti-spoofing
- Python
- PyTorch
- SQL
- Graph databases
- Fraud detection
- Deepfake detection
- Explainability
- ONNX
Qualifications
- 3+ years in machine learning with models shipped to production in fraud, risk, identity or a comparable adversarial domain.
- Hands-on GNN work with PyTorch Geometric or DGL on real, large, noisy graphs, not only benchmark datasets.
- Hands-on computer vision for face recognition or face anti-spoofing, with evaluation using FAR/FRR or APCER/BPCER.
- Strong Python, PyTorch and SQL; able to build features from raw event and identity data yourself.
- Clear written and verbal communication: you can explain why a cluster was flagged to an investigator and a risk head.
Nice to have
- Graph databases such as Neo4j or TigerGraph, and streaming graph updates.
- Fraud domains: mule accounts, synthetic identities, loan stacking, collusive borrower networks.
- Deepfake and injection-attack detection; familiarity with ISO/IEC 30107-3 presentation attack testing.
- Indian KYC flows: Aadhaar e-KYC, CKYC, DigiLocker and Video KYC.
- Learning with scarce, delayed labels: semi-supervised, anomaly detection, active learning.
- Model explainability for graph and vision models.
- Handling biometric data under the DPDP Act.
Other details
- Your fraud experience is limited to rules engines or threshold tuning.
- You have only integrated third-party face match or liveness APIs, never trained or evaluated such models.
- You have worked with graphs only on academic datasets.
Benefits
- Ownership of fraud and identity AI for a platform with over 2 Cr users.
- Real fraud labels and investigator feedback to learn from, not synthetic benchmarks.
- Direct work with the Chief of AI, Risk and Operations leadership.
- A path to lead the fraud AI function as it grows.
Company info
About LenDenClub
About VARTIS Platforms (Formerly known as LenDenClub Techserve Pvt. Ltd.) At VARTIS Platforms, we are a corporate group on a mission to democratize credit access and empower millions of people through cutting-edge technology. We focus on building trust and convenience, leveraging AI to drive the information revolution for smarter risk