Software Engineer, Fraud

Whatnot · Consumer · San Francisco, CA · Engineering

Software Engineer role focused on building and deploying ML-driven systems for fraud detection, prevention, and intervention in a livestream shopping marketplace. Responsibilities include developing intelligent user graphs, training and deploying traditional ML and LLM models, creating scalable data pipelines, real-time inference systems, and human-in-the-loop systems for continuous refinement. The role emphasizes analyzing behavioral and adversarial data to identify emerging fraud trends and evolving systems to combat them.

What you'd actually do

  1. Lead the full architecture of fraud detection, prevention, and intervention systems — spanning machine learning, backend, and client-side components.
  2. Build intelligent user graphs to model behavioral patterns, detect collusion networks, and uncover account connectivity at scale.
  3. Design, train, and deploy both traditional ML and LLM-powered models to detect fraudulent activity across users, payments, and marketplace interactions.
  4. Develop scalable data pipelines and real-time inference systems capable of supporting high-volume, low-latency ML workloads.
  5. Create human-in-the-loop systems that continuously refine detection accuracy and adapt to evolving adversarial tactics.

Skills

Required

  • 4+ years of software engineering experience building systems for consumer-scale traffic and reliability.
  • 1+ years writing production-grade Python code and working with ML libraries (e.g. PyTorch, LightGBM).
  • 1+ years of experience in machine learning or fraud prevention domains.
  • Fluency with data tooling, including data warehouses (e.g. Snowflake) and transformation frameworks (e.g. dbt, Dagster).

Nice to have

  • Hands-on machine learning or data science experience in production environments.

What the JD emphasized

  • ML-driven systems
  • real-time detection
  • LLM-powered models
  • low-latency ML workloads
  • human-in-the-loop systems
  • emerging fraud trends
  • adaptive, production-ready systems

Other signals

  • ML-driven systems
  • real-time detection
  • user-centered enforcement
  • behavioral patterns
  • collusion networks
  • account connectivity
  • LLM-powered models
  • fraudulent activity
  • high-volume, low-latency ML workloads
  • human-in-the-loop systems
  • emerging fraud trends
  • adaptive, production-ready systems