Portfolio — Applied AI

Duromex

Every appliance, right the first time.

Seed 2026 · $1.5M Applied AI London, United Kingdom

Duromex runs the appliance line from sheet metal to packed unit: forming, stamping and enamel control, AI surface, weld and assembly-defect vision, polyurethane foaming and insulation control, wiring and component assembly checks, refrigerant and water leak and functional-test intelligence — all on a line-and-product digital twin that predicts yield and energy rating before the build.

Appliances are a thin-margin business where a foam void turns into a failed energy rating and a half-seated connector turns into a warranty claim eighteen months later. The data to catch both already exists on the line; nothing acts on it in time. Kabeer and Maria are tying the end-of-line test result back to the press stroke and the foam shot that caused it, which is the part nobody else has built.

Duromex

  • Seed · 2026Stage
  • $1.5MRound
  • Applied AISector
  • London, United KingdomHeadquarters
  • 11Team
The team

Kabeer Iqbal

Co-founder & CEO

Kabeer runs the commercial side and the product direction: which line goes first, how the success metric is written down, and who at the OEM signs off on it. He started Duromex because appliance plants are full of test data that never changes what happens at the press or the foamer.

Maria Li

CTO

Maria leads engineering: edge vision and sensor fusion that decide within takt, the line-and-product twin, and the bounded, revocable write-back into line controllers. She is why every agent earns its authority in shadow mode first, with an evidence trail behind each action.

How it's gone
  • 2026

    We came in

    $1.5M seed, with design partners across refrigeration and laundry lines.

  • 2026

    Shadow mode

    Baselines running against live foaming, assembly and end-of-line test data before any write authority.

  • 2026

    Building the team

    Hiring edge-AI and controls engineers in London.

An illustrative 99SA Ventures profile — founder, team and figures stand in until the company shares its own.

Building AI that has to be right, not just impressive?

If your model gets graded against something real, we're the easy meeting to get.