The Fifth Elephant
The Fifth Elephant

The Fifth Elephant 2026 Winter Edition

Call for Data Problems and Proposals

All submissions
Anoop B S

Anoop B S

@anoopbs

Pawan Kumar Singh

@iampawansingh Author

Real-time Feature Store for Fraud and Risk Management

Submitted Oct 12, 2026

🎤 I am submitting: To speak I have a submission for: A talk - 30-40 mins

Submission type: Talk/session proposal — 30–40 minutes

Session

Building and deploying real-time features for lending risk and fraud models remains slow and operationally complex. A feature definition that works during model training must be made available reliably at inference time, but in a very quick time-to-market, while accounting for external bureau and vendor data dependencies, latency, governance and parity between training and production.

We built a Feature Store SDK to make this path from training to production close to “click-and-go.” In this session, we will share the real-world design choices behind the platform, the alternatives we explored, and how we arrived at our current approach. We will also discuss how the solution enables us to ship tens of models and hundreds of new production features each quarter for real-time Risk and Fraud use cases.

Takeaways

  • Understand why real-time Fraud and Risk features in lending are more complex than traditional feature engineering, particularly when external data, latency, reliability and speed matter.
  • Learn the options we evaluated, the trade-offs that shaped our final architecture, and how a write-once-use-everywhere Feature SDK can make feature deployment repeatable at scale.
  • Learn how the Feature SDK allow defining features that can run as Pandas, Spark or PySpark pipelines without any change in underlying definition.

Audiences

ML Engineers and Applied Data Scientists working in Fintech will benefit most. Practitioners from other industries will also find the session useful, especially those building real-time ML systems beyond the more familiar social media, media streaming and e-commerce feature-engineering patterns.

Bio

Anoop B S is a Product Manager, and Pawan Kumar Singh is an ML Architect, for the ML Platform at Oportun - a US financial services company that lends to underserved communities and customers with thin credit files.

Before Oportun, Anoop and Pawan built ML platforms at Rakuten and BetterPlace. At Oportun, they are building the platform capabilities that help Data Science teams move real-time Risk and Fraud features and models from development into production reliably and at speed.

Draft slides

Not yet prepared. We are currently creating an external-facing deck and will share it before the editorial call. The session will use concrete architecture, trade-off, and production-scale examples.

What we do not know yet

We welcome critique of our approach, particularly around architectural choices and scalability challenges that industry practitioners foresee as real-time feature platforms scale across more models, teams and data sources.

Match-making tags

We can help with: ideas, experience, technique, critique.
We need help with: critique.
We’d like to meet: practitioners, domain experts, tool builders.

Topic tags

fraud, underwriting, financial services

All submissions

Comments

{{ gettext('Login to leave a comment') }}

{{ gettext('Post a comment…') }}
{{ gettext('New comment') }}
{{ formTitle }}

{{ errorMsg }}

{{ gettext('No comments posted yet') }}

Hosted by

The Fifth Elephant
The Fifth Elephant
Jumpstart better data engineering and AI futures