We are filing an AI or machine learning invention in India. What should the specification contain?

The 2025 Guidelines frame sufficiency as two questions: what the invention is, and how it is implemented.

For the "what", the specification is expected to describe each feature, with drawings for hardware-based inventions and flowcharts or equivalent for methods, together with the relationships and connectivity between components and the output or intermediate results produced. For the "how", the applicant is expected to disclose the best known method of performing the invention.

In practice, for an AI or ML invention this means going beyond naming a model architecture. Set out the training and inference steps, the data flow, the interaction with hardware, and — most importantly — articulate the technical problem being solved and the technical effect achieved, in the specification itself rather than for the first time in response to a First Examination Report. Applications drafted for another jurisdiction and filed unchanged in India are where Section 3(k) objections tend to become difficult to answer.

Source code is not required.