Senior Machine Learning Engineer

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 card price move" is a question, not a query

Use Claude Code or Cursor as the default way you work through a refactor or an infra migration, not as autocomplete

If you've built agents, worked with the API directly, written skills or MCP servers, or shipped internal tooling your team depends on — lead with that

7+ years engineering, with 5+ years building and shipping production ML/AI models

Production-grade Python and SQL , including custom feature-engineering pipelines — not just off-the-shelf scikit-learn. Time-decay weighting, leakage-safe k-fold cross-validation, cascading fallback and imputation logic

Gradient-boosted or ensemble estimators trained and validated against strict accuracy tolerances, with segment-specific tuning by category or asset type

LLMs and foundation models in production , for both internal tooling and user-facing product

MLflow or a comparable tool for experiment tracking and model registry/versioning

Production model-serving on AWS that was yours — capacity planning, autoscaling, and diagnosing memory and timeout failures at scale

CI/CD, production workflow orchestration, and IaC

A pragmatic bias. Value delivered incrementally over the perfect rebuild

Bonus: real-time or low-latency serving at scale. Startup experience. You collect, or you have opinions about the collectibles market.

We'll be specific so you can prepare. Before any of it: download the app, buy, bid, and sell. Explore our industry. Every round assumes you've used the product, and we expect your point of view to sharpen as you go.

1. Recruiter screen — 25 minutes. Come ready to talk about the metric you own today, why Alt specifically, and how you're actually using AI. We'll answer your questions and align on comp.

2. Hiring manager with Dae, Head of Data — 45 minutes. Be ready to go deep on a model you took to production and kept there: the systems design, what you owned, where it broke, and what you'd do differently.

3. System design— 45 minutes. Design a model-serving path under real constraints. We'll push back on at least one of your choices on purpose — how you take that matters as much as the design.

4. Team interview- 45 minutes. How you work with the people who depend on your models, including our Expert Pricers.

5. Dae — 30 minutes , then Leore, our founder and CEO — 30 minutes. Everyone who joins Alt meets Leore. Come with a point of view on where you'd take our pricing intelligence.