Reinforcement learning · Credit risk · Capital One
You can't experiment with credit policy on real applicants — so I built a simulation engine that generates synthetic loan applications to power a reinforcement-learning task that optimizes policy offline.
This is a deliberately high-level account of proprietary work at Capital One — the problem, my role, and the approach, described qualitatively. It contains no confidential figures, internal systems, data, models, code, vendor or team names, or architecture.
The idea
Credit policy is a sequential decision problem well suited to reinforcement learning — but the one thing you cannot do is learn on live applicants. The way through is a faithful simulator: generate synthetic applications, let an RL agent optimize policy against them, and evaluate decisions before any of them touch a real person.
What I built
A simulation engine that produces synthetic loan applications, and the reinforcement-learning task that optimizes credit policy against that simulated population — a safe, offline testbed for policy decisions.
What it demonstrates
Applying reinforcement learning and simulation to a safety-critical decision system: getting the benefits of learned policy optimization while keeping experimentation entirely off real customers.