NeoBram expert profile

Karthick Raju, Chief of AI.

Karthick Raju leads NeoBram's industrial AI practice across architecture, private deployment, production evaluation, capability transfer and AI delivery enablement.

Discuss an industrial AI problem

Subject areas

Where Karthick represents NeoBram's AI practice.

Discuss one workflow, bottleneck or customer opportunity and shape a practical path from discovery to production.

Industrial AI architecture and production evaluation

Private, offline, on-premises and edge deployment

Source-grounded knowledge assistants and retrieval evaluation

AI operating models, capability transfer and delivery enablement

Manufacturing, pharma, oil and gas, and EPC use-case qualification

Working positions

How these engagements are approached.

Architecture follows the operating decision

Model and platform selection come last, not first. The starting questions are which decision the system supports, what evidence a person currently trusts to make it, what happens when the system is wrong, and who is accountable after go-live. An architecture chosen before those answers exist usually has to be rebuilt.

Private deployment means more than on-premises hosting

A credible private deployment has to account for identity, access control, logging, model and dependency updates, backup, support and licence rights for every component in the stack. Offline, on-premises, edge and private-cloud each trade differently against those, and the right boundary depends on which data actually cannot leave.

Retrieval systems are judged on traceability, not fluency

For industrial knowledge assistants, the useful measure is whether an answer can be traced to a controlled source and version, and whether the system declines and escalates when evidence is missing. A confident answer from a superseded document is a worse outcome than no answer, because it will be acted on.

Capability transfer is a delivery requirement

Engagements are structured so the client team can operate, evaluate and change the system afterwards. That means documented intended use, an evaluation method the team can re-run, named owners and an explicit transition point — rather than an indefinite dependency on the people who built it.

Use-case qualification includes the reasons to stop

A qualification that can only conclude “proceed” is not a qualification. Manufacturing, pharma, oil and gas and EPC workflows are assessed against data availability, integration boundary, operating risk and business value together, and a documented reason to close a gap, pick another use case or stop is a legitimate outcome.