CPT Work Authorization for Graduate AI and Data Science Engineers

The curricular test comes first

Curricular Practical Training exists to support work that is an integral part of an established curriculum. That framing drives everything else. Before an offer is discussed, the placement must map onto a course, a required practicum, a thesis or a programme requirement that the designated school official can point to in the catalogue.

For graduate AI and data science engineers this mapping is usually straightforward but rarely written down well. A machine learning practicum that requires applied work on real data pairs naturally with an engagement building a feature pipeline or an evaluation harness. The weakness is almost never the substance — it is that nobody recorded the link between the syllabus and the scope of work.

Sequencing the authorisation correctly

CPT is authorised by the designated school official, recorded on the student's immigration record, and must be in place before the first day of work. Employment that begins before authorisation appears cannot be retroactively cured, and it can jeopardise later filings, including a STEM OPT extension or an H-1B petition. Employers who hire graduate AI talent at volume should therefore build authorisation confirmation into onboarding as a gate, not a formality.

• Confirm the authorisation dates before the start date is agreed.
• Match the employer name and location on the record to the actual worksite.
• Re-authorise for each academic term rather than assuming continuity.
• Keep the syllabus or programme requirement on file with the scope of work.

Part-time, full-time and the twelve-month threshold

During the academic term CPT is generally part-time at twenty hours per week or less, while full-time CPT is typically reserved for vacation periods or programmes that require it. The distinction is not cosmetic: twelve months or more of full-time CPT eliminates eligibility for optional practical training at the same degree level. Teams that plan a co-op sequence should model the cumulative total early, because the consequence surfaces long after the scheduling decision was made.

Practically, this argues for careful timesheet discipline. Where a graduate engineer supports a production MLOps rotation, define the hours contractually and monitor them, rather than letting an intense release week quietly convert a part-time placement into a full-time one.

Designing scope that survives scrutiny

The strongest CPT engagements read like supervised applied research. Give the engineer a bounded problem with an academic dimension — data drift measurement, retrieval evaluation, causal inference on an observational dataset — and require a written deliverable that the academic supervisor can assess. That deliverable simultaneously satisfies the curricular requirement, gives the employer durable value, and creates the documentary record that later filings depend on.

DeltaDex Technologies structures graduate placements this way as a matter of course, pairing each engineer with a supervising researcher and a defined evaluation artefact so the academic and commercial objectives point in the same direction.