AI drafts.
Attorneys file.

The boutique handles H-1B, O-1A, and EB-1A/EB-2 NIW petitions for engineering and research talent moving between India and the US. Each petition is a long, evidence-heavy assembly: experience letters, reference letters, publications, conference talks, patents, peer-review records.

The bottleneck was never legal judgment. It was reading-and-summarizing time — turning forty supporting documents into a coherent narrative section that an attorney could then refine.

Supporting documents attach to a Petition matter and are tagged by evidentiary category (extraordinary-ability criteria, employment evidence, publication record). Clara’s assistance layer drafts a summary section per category, citing the underlying documents inline.

Critically, every drafted section enters the matter as Status: AI Draft — Pending Review. The petition cannot progress until an attorney explicitly signs off on each section. The signature, the model version, and the document references are stored on the matter as audit metadata.

Median time from intake-complete to first-draft-ready dropped from sixteen business days to under five. Attorney time per petition is now concentrated on framing and judgment calls, not on summarizing letters.

The audit metadata on every petition has also become a regulatory comfort: when the firm is asked how AI is used, it can show exactly which sections were AI-drafted, which attorney reviewed them, and what version of the model produced them.