Capital One·Machine Learning Engineer·Onsite - System Design / Architecture·Senior

SeniorPrefer not to say

Jun 2026

Summary

System design round at Capital One for an ML Engineer role, and the prompt was way more fintech-specific than I expected. The whole session revolved around virtual credit card infrastructure, which felt more like a payments engineering problem than anything ML-adjacent.

Questions Asked (1)

Design a virtual credit card system that issues single-use or merchant-locked card numbers on demand, backed by a real funding source. Walk through the API design, data model, authorization flow, fraud safeguards, and non-functional requirements.

System DesignAPI & IntegrationsData Modeling

Author's notes

This was a beast of a question and I didn't scope it well at the start.

Create a free account to read the full note

AI HintsAI Generated

Suggested Approach

Start by clarifying requirements and scope, then structure your answer around the five requested areas: API design, data model, authorization flow, fraud safeguards, and non-functional requirements. Emphasize how ML models can be integrated into fraud detection and authorization decisions, given the ML Engineer role.

Pro tip: Highlight the trade-offs between security, latency, and user experience, and discuss how ML can be used for real-time fraud scoring without adding significant latency.

1. Clarify Requirements and Scope

Ask questions to understand expected scale, latency requirements, regulatory constraints, and integration points with existing systems. Define the core features: single-use and merchant-locked cards, funding source management, and real-time authorization.

2. Design APIs and Data Model

Outline RESTful APIs for card issuance, funding source linking, and transaction authorization. Define key entities like VirtualCard, FundingSource, Merchant, and Transaction, with relationships and attributes.

3. Detail Authorization Flow

Describe the step-by-step process when a transaction is attempted: card validation, merchant lock check, funding source verification, fraud scoring, and authorization decision. Include how ML models fit into the flow.

4. Incorporate Fraud Safeguards

Discuss fraud prevention techniques such as velocity checks, anomaly detection, and ML-based risk scoring. Explain how models are trained, deployed, and monitored for real-time decisions.

5. Address Non-Functional Requirements

Cover scalability, availability, latency, security, and compliance. Explain how the system handles high throughput, ensures data consistency, and meets regulatory standards like PCI DSS.

Key Points to Mention

API design: endpoints for card creation, funding source management, and authorization with idempotency and rate limiting.
Data model: entities like VirtualCard (with attributes: card number, expiry, CVV, merchant lock, single-use flag), FundingSource, Transaction, and Merchant.
Authorization flow: real-time checks including card status, merchant match, funding source balance, and ML-based fraud score.
Fraud safeguards: ML models for anomaly detection, velocity rules, and integration with third-party fraud services.
Non-functional requirements: low latency (<100ms), high availability (99.99%), scalability, PCI DSS compliance, and audit logging.
ML integration: feature engineering, model serving (e.g., TensorFlow Serving), A/B testing, and monitoring for drift.

AI-generated suggestions, not part of the candidate's original notes. May be inaccurate — verify before relying on them.