VigilOne · our immigration-law model

A model trained to reason like an immigration lawyer.

Ask a general-purpose AI an immigration question and it answers with confidence, even when the rule it gives you does not exist. VigilOne is built for one field and one job. It reads a client's facts the way a specialist does, finds the law that governs them, and answers from that law alone.

How it thinks

The answer shape, every time.

  1. Spot the issues. A fact pattern rarely asks one question. VigilOne names each issue before it answers any of them.
  2. Find the governing law, in order. Statute, then regulation, then agency guidance, then decisions: the order the law binds.
  3. Answer, cite, date. Every statement carries the provision it rests on and the date that provision was read.
  4. Say what is left open. When the facts or the law do not settle a point, the answer says so, and says what would settle it.
VigilOneReasoning, shown
"My H-1B employer laid me off last week. How long can I stay, and what are my options?"
Issues spotted
Grace period after H-1B employment endsMoving to a new employerChanging status
Governing law, in order
StatuteINA § 214(n)H-1B portability to a new employer
Regulation8 CFR § 214.1(l)(2)Up to 60 days, or to the end of the validity period if shorter
Regulation8 CFR § 204.5(p)Compelling-circumstances EAD, in narrow cases with an approved I-140
Answer

Up to 60 consecutive days to stay in the U.S. without H-1B employment, or until the I-94 ends if sooner. Within that time a new employer may file for you, you may change to another status, or you may leave.

What the facts don't settle

The 60 days are available once per validity period. Whether they were used before is not in the facts.

Example of the answer shape VigilOne is trained to give. Not a released answer.
How it is trained

Taught by a specialist. Graded like a lawyer.

Retrieval-grounded

Every training answer is built from primary-law excerpts: the INA, the CFR, the agency manuals and precedent decisions.

Supervised fine-tuning (LoRA)

On answers written and corrected by a board-certified immigration specialist.* Each correction becomes the next lesson.

Preference optimization (DPO)

Taught to prefer the answer that stays inside its sources over one that sounds right and is not.

Reinforcement learning (GRPO)

Verifiable rewards: an answer earns nothing if a cited section does not exist or a quotation does not match the source.

Held-out blind evaluation

Each version sits an exam it has never seen. One invented rule fails it.

Private

Trained and served on VisaVigilAI's own infrastructure.

Where it stands

In training. It ships when it passes, and not before.

Until then, Vigil inside CaseVigil answers with a general model that reads the same law library and follows the same rules.

The library it reads
  • The Immigration and Nationality Act and related titles of the U.S. Code
  • Titles 8, 20, 22, 28, 29, 42 and 45 of the Code of Federal Regulations
  • The USCIS Policy Manual
  • The Foreign Affairs Manual
  • The EOIR Policy Manual
  • USCIS form instructions
  • AAO, BALCA and federal court decisions
  • DOL OFLC guidance

Each source is re-read on a schedule, and every answer carries the date it was read.

Talk to the lawyers who build it.

No sales team. Write to the founders and we will show you the platform running on a real practice.

Request a demo rajesh@visavigilai.com bindi@visavigilai.com