Sino — industrial AI, from zero to customer sale

At a glance

• Industrial clients set hard cost, latency and hardware ceilings; accuracy that cannot run inside them is irrelevant.
• Industrial AI products across language models, graph neural networks and computer vision, optimised to fit each client.
• Completed: products taken from zero to customer sale over three years, from May 2022 to May 2025.

Highlights

• Language, graph and vision models

  Large language models, graph neural networks and computer vision, applied in industrial products rather than only in research.

• Fitted to cost and latency ceilings

  Models optimised to run within the client's cost, latency and hardware limits, not just in the lab.

• From zero to a sold product

  Products carried from the first idea through delivery and sale to the customer.

The challenge

Industrial clients work under hard ceilings on cost, latency and hardware. Research-grade accuracy is irrelevant if the model cannot run inside those limits.

That constraint is the difference between a demo and a product a factory will actually pay for. The engineering question was never only whether a model works, but whether it works on the client's hardware, within the client's latency limit and at a cost the client accepts.

Architecture

Several products across large language models, graph neural networks and computer vision. The models were systematically optimised to fit each client's cost, latency and hardware ceilings, and were analysed with proposals for improving them.

What I built

Industrial AI products were delivered from zero to customer sale. The models behind them were systematically optimised to fit industrial and cost constraints.

• Frontier and industrial problems solved with AI.

• Products taken from zero to sale, through delivery to the customer.

• Constraint-fitted models, optimised for each client's industrial and cost limits.

• Model analysis and improvement, with systematic proposals for making each model better.

• Oversight of external delivery: large projects built by other teams were directed, reviewed and signed off.

Outcome

Industrial AI products reached customer sale, and the engagement closed in May 2025 after three years, Mohammad's longest tenure. Alongside his own delivery, he directed, reviewed and signed off large projects built by other external teams.

My role

• Took industrial AI products from zero to a delivered product sold to customers.
• Optimised AI solutions to fit each client's industrial and cost constraints.
• Applied large language models, graph neural networks, computer vision and deep learning to frontier and industrial problems.
• Analysed AI models and proposed systematic improvements to them.
• Directed, reviewed and signed off large projects delivered by other external teams.

Engineering

• Constraints are requirements

  Cost, latency and hardware ceilings are treated as requirements: accuracy only counts if the model runs inside them.

• Systematic optimisation

  Models were systematically optimised to fit the client's industrial and cost constraints.

• Review and sign-off of external work

  Large projects built by other external teams were directed, reviewed and signed off.

Outcomes

• Industrial AI products reached customer sale.
• Three years of delivery, May 2022 to May 2025: Mohammad's longest tenure.
• Large projects built by other external teams were directed, reviewed and signed off.