H-1B Specialty Occupation Proof for Deep Learning and NLP Specialists

What the specialty occupation standard asks

An H-1B petition must show that the position requires theoretical and practical application of a body of highly specialised knowledge, and that the minimum entry requirement is a bachelor's degree or higher in a specific specialty. Deep learning and natural language processing roles usually satisfy this comfortably in substance, yet petitions still draw requests for evidence because the paperwork describes the work in generic software terms.

The failure mode is predictable. A description that lists "develops software features and collaborates with stakeholders" invites the conclusion that a general degree, or no degree, would suffice. A description that requires probability theory, optimisation, transformer architecture internals and evaluation design makes the degree nexus self-evident.

Writing duties that carry the degree nexus

Each duty should name the specialised knowledge it consumes and, where possible, the proportion of time allocated to it. Reviewers respond to specificity: fine-tuning objectives and loss function selection, tokenisation and subword vocabulary decisions, retrieval evaluation with ranking metrics, quantisation trade-offs measured against perplexity or task accuracy, and statistical significance testing across evaluation runs.

• Allocate percentages of time across duties so the role's composition is visible.
• Name the mathematical and computational foundations each duty requires.
• Reference the specific degree fields that supply that foundation.
• Avoid interchangeable phrasing that would fit any engineering role.

Supporting evidence beyond the job description

Strong petitions layer several kinds of proof. Industry evidence shows that comparable organisations require a specialised degree for parallel roles: job postings from peer firms, professional association materials, and salary surveys that classify the occupation. Internal evidence shows consistency: an organisational chart, prior hires in the same role with the same degree requirement, and a written hiring standard.

Expert opinion letters can help where the role is genuinely novel, but they work best as corroboration of a well-drafted description rather than as a substitute for one. The same is true of the trainee's own record — publications, conference presentations, released models and patents demonstrate that the work sits in a research discipline.

Common pitfalls

Three problems recur. First, a wage level that contradicts the seniority claimed in the description; the two documents must tell the same story. Second, a generic occupational classification that undersells a research role. Third, third-party placement arrangements without a clear chain of contracts and end-client documentation establishing who directs the work.

Resolving these before filing is far cheaper than answering a request for evidence later. The underlying principle is simple: the petition should describe the role the way the engineering organisation actually describes it internally, with the mathematics left in.