An AI first-pass reviewer for PJM interconnection submissions

The most expensive email in data center predevelopment is the deficiency letter. You submit the feasibility study or the System Impact Study application, four weeks later a paragraph comes back asking for missing reactive capability data, an unclear POI designation, or a gross-versus-net MW mismatch, and the project slides a cycle. Most of those deficiencies are catchable before submission. Almost none are caught today.

The Cliff submission validator reads your application against the PJM tariff and the study manuals before you send it. It returns a readiness score, a ranked list of findings at four severities, the specific tariff citation behind each finding, and a recommended fix, sometimes including an auto-patch. Ten seconds of LLM inference replacing four weeks of PJM review-queue time.

This only works because frontier models can finally read a 180-page tariff, a 60-page study manual, and a 40-field application form together in a single context window and reason about the interactions. A 2020-era rule engine could catch missing fields. It could not tell you that your reactive capability curve is inconsistent with your claimed fuel type, or that your requested in-service date is infeasible given your queue window. That is the difference.

Start on the sample submission to see what the output looks like, or bring your own at /submission-validator/run.