Senior Machine Learning Engineer

Alt is unlocking the value of alternative assets, starting with the $5 B trading-card market. We let collectors buy, sell, vault, and finance their cards in one place and we are backed by leaders at Stripe, Coinbase, Seven Seven Six, and pro athletes like Tom Brady and Giannis Antetokounmpo. Our next frontier is real-time pricing at scale—the Alt Value that powers every trade, loan, and product on the platform.

Every buyer, seller, and lender on Alt is acting on a number our models produced. Alt Value prices the card. The underwriting model sizes the advance. Those two systems are the difference between a marketplace and a handshake, and they are the closest thing we have to a moat.

We’ve proven model-driven pricing works. This role takes it from working to excellent: more coverage, better accuracy, lower cost to run, faster to refresh. You’ll own the full lifecycle — feature generation, training, validation, deployment, serving, and the monitoring that catches drift before a customer does.

This is not a research seat. The models exist. What they need is someone who treats production as the deliverable.

The metric you own:
Model-Based Pricing Coverage — the percentage of cards confidently priced by models rather than by hand. Supporting KPIs: pricing accuracy (% error), pricing freshness (end-to-end orchestration time), and underwriting performance (advance disbursement rate against target default rate).

Most people who are great at this have owned a model in production where being wrong cost money — pricing, risk, credit, or fraud — not a notebook that got handed to someone else to deploy.

• Leaner, sharper pricing. Cut infrastructure cost meaningfully while improving accuracy, especially on high-value assets where being wrong is expensive.

• Underwriting from good to great. Iterate the model to maximize cash advance disbursements without breaching risk thresholds or default targets.

• The full ML lifecycle. Feature generation and training through deployment and monitoring. No handoff, no throwing it over a wall.

• The production path. The models’ AWS infrastructure and the pricing APIs themselves — capacity planning, autoscaling, latency, and diagnosing the memory and timeout failures that only show up at scale.

• Experiments that settle arguments. Design and run backtests to find and validate features that actually move predictive power and coverage.

• Domain depth. Work directly with our Expert Pricers until your model changes reflect real market judgment, not just what the data allows.

• Shipped leaner, more accurate pricing models. Infrastructure cost is meaningfully down and accuracy is up, especially on high-value assets.

• Moved underwriting from good to great. More disbursement, no breach of risk thresholds.

• Earned trust with Expert Pricers. You’re their go-to partner, and they can tell your changes reflect the market.

• Hardened the production path. The pricing APIs are faster, more observable, and easier to reason about, with drift monitoring in place before it reaches customers.

We don’t want to hear that you use Claude every day. Everyone does. We want to know what you’ve built with it. Concretely, in this role:

• Build an agent that reads a batch of new comps and flags the ones our model is likely to price badly , before a pricer has to catch it by hand

• Use foundation models for feature extraction from unstructured card and listing data — condition language, provenance notes, auction descriptions — and get real lift out of it

• Stand up a backtest harness you can talk to , so evaluating a feature idea is a conversation rather than a two-day branch

• Put an MCP server over the model registry and prediction logs so “why did this…