<aside> 🗣

Interested? Email us at [email protected] with your resume and anything you've built that you're proud of — a repo, a paper, a demo, a write-up.

</aside>

About Vardera

Vardera builds the intelligence layer for collectible commerce: computer-vision infrastructure for any category where a product image needs to be identified, graded, valued, or screened for fraud. Four capabilities, delivered as a single API — Identify → Grade → Value → Trust. Our category-specific vision models replace the manual grading-house workflow (slow, expensive, expert-dependent) with sub-second, sub-dollar inference.

We're live in production on coins, trading cards, and action figures, and we sell to enterprise marketplaces, large retailers, vertical platforms, and brands. Founded in 2025, seed-stage, ~$5.2M raised from Animo VC, Garuda Ventures, Analog Ventures, Rackhouse VC, and eBay Ventures.

This Is an Open Application

We don't wait for a req to open to meet good people. We are always hiring AI and ML engineers with real computer vision experience, at every level from strong early-career to staff.

If that's you, apply. Don't worry about whether there's a posted role matching your title, your years of experience, or your exact stack. If you can build vision systems that hold up on messy real-world images, we want to talk. We'll figure out the right role and level together.

What You'd Work On

Specifics depend on where you land, but the work looks like this:

• Train and deploy category-specific vision models that identify, grade, value, and screen physical goods — down to the level of a specific die variety on a Morgan Dollar, not just "a coin."
• Push accuracy where it actually matters economically. The difference between an 1878-CC Morgan in MS-67 and the same coin in MS-63 is roughly $65,000 and $750. A general-purpose vision model cannot tell them apart. Ours has to.
• Own the full loop: training-data acquisition and annotation strategy, model development, evaluation, deployment, and the feedback pipeline that makes the next version better.
• Define and evolve accuracy and quality metrics beyond standard loss functions — optimizing for the real-world economics of error and trust.
• Build evaluation and monitoring pipelines that track performance by category, lighting condition, and defect type.
• Hold the line on latency and cost. Most responses run 1–2 seconds end to end at sub-dollar cost, and that's a product constraint, not a nice-to-have.
• Take a new category from nothing to production, and work directly with founders and enterprise customers to turn ambiguous problems into measurable, repeatable results.