AI for immigration and visa applications, from applicant questions to case preparation

Two deployed USCIS systems prepare immigration applications for officers. PDF Intake lets people submit scanned forms online through myUSCIS instead of by mail: a generative model on Amazon Bedrock extracts the filled fields and checks them against form specific business rules, returning structured JSON to the case system. The ELIS Evidence Classifier tags scanned evidence pages (such as a marriage certificate or a border crossing card) so that adjudicators can jump to the document they need instead of scrolling through hundreds of pages. Their outputs are structured form data and evidence tags for the case system.

Consular Affairs is piloting Live Consular AI Language Augmentation (LCALA): real time transcription and neural machine translation of the spoken exchange at the visa interview window, with translated audio and on screen text and an optional short time stamped transcript. The visa interview pilot is flagged high impact in the federal inventory; its entry notes that interviews last about three minutes and that misunderstandings can force repeat questions, delays or uneven outcomes. A second pilot, for Overseas Citizens Services and American Citizens Services, supports calls and in person interactions and is described as assistive only, not a replacement for certified interpreters where they are required.

The Student and Exchange Visitor Program response center, which serves international students and school officials, is preparing SID (SEVIS Interactive Dialog), a voice assistant that answers routine caller questions through a deterministic question and answer workflow. When it cannot answer, it turns the caller over to an agent in the response center and creates a ticket with a caller transcript to reduce the burden on the agent. Separately, it captures the interaction and sends it through an API to the program's automated management system (SEVPAMS).

Since 25 April 2023 the Complexity Application Routing Solution for visits (CARS(V)) labels visitor visa applications under Appendix V as likely non complex or complex, so each goes to the right grade of decision maker, using business rules on the applicant's declared answers (such as a yes to any criminality question) plus matches against risk profiles and bulk data tables of information previously seen in applications that gave false information.

It uses no machine learning, does not decide applications, and decision makers can reroute a case; its equality impact assessments allow direct discrimination on race or nationality where a Ministerial Authorisation supports it, new profiles are applied weekly, and the profiles and bulk tables are not set out in the record.

The record gives around 2.5 million visit applications received in 2023 for scale, not a count routed by the tool; separately, the Home Office withdrew an earlier visa streaming algorithm in 2020 after a legal challenge.

Since 2018 IRCC has used an advanced data analytics system, built by its own data scientists, to sort temporary resident visa applications. It first covered applications from China and India and, from January 2022, every visitor visa application submitted from outside Canada. The system flags routine files, such as those from applicants previously approved to visit Canada in the past 10 years, for faster handling, and triages other files to officers by complexity.

It can find an applicant eligible but never refuses or recommends refusing an application, and a share of the files it clears is sent to full human review as a fairness check. Officers make every final decision, and IRCC published an algorithmic impact assessment that rates the system moderate impact.

The Immigration and Citizenship Services Authority (ICSA) of Papua New Guinea, with its AWS partner NiuPay, built the MyICSA Analysis Engine (MAE) into its eVisa platform. MAE reads each short term visa application with Amazon Bedrock for text, Amazon Textract for scanned passports and letters, and Amazon Rekognition for photo validation, then checks the documents against configurable rules that immigration officers maintain without technical help. NiuPay's chief technology officer says the engine grants a visa automatically only when it is 100 percent sure and escalates any uncertain application to a human reviewer. The system went live after testing on 100,000 sample cases, and ICSA plans to extend it to work permits and passports.

The General Directorate of Residency and Foreigners Affairs in Dubai (GDRFA) built the Entry Permit 50+ project, an automated system that uses artificial intelligence to scan applicants' documents and issue electronic entry permits, including visit visas and renewals, without human intervention. Applicants apply through the GDRFA website and smart app, and the system is linked to other government bodies so security checks run in the same step. Officials said at a press conference that scanning and issuing a permit, which took six hours, now takes 15 seconds when the applicant supplies the necessary documents.