Home
← Back

AI Robotics Startup · Oct 2024 – Sept 2026

Gradient

Robots, vision, data, and ML infrastructure for high-volume card handling.

Work

My top selected projects over 2 years at Gradient. I led the pivot to a robotics-manufactured product line after building most of the robotics and data layer underneath it. I’m not just an individual contributor, but a strong leader of engineers as well.

Leading the Website Buildout

Our old site was failing under millions of SKUs. I led the design and build of its replacement.

  • $102Kin 7 weeks
  • 4,055paid orders

Leading the pivot

Out of consignment and into repacks, with three months of runway left.

  • $2M/yrlosses on the old model
  • 20%margin after pivot

Demoing to eBay

Presented the robotics division to eBay's CFO, Head of Collectibles USA, and four other senior executives.

Sorting robot

The machine the pivot depended on. Sorts on price, team, player, or anything else the catalogue knows.

  • 1,500cards / hr

Building the Data Backbone

Forty Docker nodes, a data lake, and a parser. The whole company's data operation.

  • 40nodes in parallel
  • 20-100/spages scraped

The Pricing Database

This simple yet reliable system was the foundation for all pricing at Gradient.

  • 40Mrows

Sync architecture

Postcard: a template for ETL nodes that moved data between every system we ran.

  • 1000/son some nodes
  • 99.9%+uptime

Scanning robots

Eight machines imaging cards at 520 ppi.

  • 2,000cards / hr, one machine
  • 520 ppicapture

The LLM Identification Pipeline

Identifies a card from its scan against a catalogue of millions.

  • 99.98%accuracy on Pokémon
  • 4.1/squeries

Inventory system

The system of record for every card in the building, down to its position in a stack.

  • 2.5Mcards per aisle

Calibration and cropping

Aligning two cameras to the scanner frame, and the classical cropper built on top of it.

  • 95%crop success
  • 0.2sinference

Prototype machines

Three pick-and-place robots I built to find the patterns worth committing to.

  • 3robots
  • 6 weeksstart to finish
← All work