Home/Team
Team
Cassation is built by lawyers who run cases themselves and by engineers who know how to build artificial intelligence systems. This is not "lawyers who had software written for them", nor "developers who read up on courts": two competences meet in this team, and the product grew at that junction.
We started from our own practice: the same volumes, the same deadlines, the same thousands of pages to read before a hearing. We build the platform for advocates, lawyers and in-house legal teams — for those who answer for their conclusions personally and cannot afford to cite something that is not in the case file.
Why now
Models change faster than the legal industry can verify how reliable they are. Every few months a model appears that writes more convincingly than the last — and almost as regularly a new reason appears not to trust its words about a case.
The industry cannot keep up: tools appear faster than anyone can make sure they are trustworthy in work where the cost of an error is not the reputation of a chatbot but the position in a case.
We build the platform so that this race does not concern us. Models are replaceable: one today, another tomorrow, and swapping is a setting rather than a rewrite. The verification is ours, and it does not depend on which model happens to be best right now.
Approach
The usual approach is one request to a model and one answer. We do not treat such an answer as a basis for a conclusion: nothing but the model itself has verified it.
It does not miss what is missed because of sheer volume, and it checks what nobody checks. This is not a slogan but something measurable on a benchmark set of cases: how many contradictions the system found, what manual review missed, and how many statements were left without a citation.
A statement does not enter a conclusion until it has been checked independently: an independent reviewer looks for places where the conclusion does not follow from the materials, the opposing side's position is assembled separately, and an arbiter rules on every point.
the arbiter is always a different model from the one that wrote the answer
Materials become a structure — persons, links, facts, events, contradictions — and every fact is tied to a specific fragment: volume, page, paragraph.
our own technology, from extraction to links
Discrepancies are sought between documents and volumes rather than inside a single text: the system stitches together testimony, expert reports and documents and shows where they contradict each other.
the final arbiter is required to re-check its own ruling
The lawyer gets a working picture: what is established, what is disputed, where the position is weak, which deadlines are running and what has to be prepared.
the output is a decision, not a text
Boundaries
Who it is for
For those who run cases alone or in a small team and answer for their conclusions personally.
For those who value the confidentiality of materials and predictable spending on case work.
For those who need their own perimeter, a contract and a clear data-processing procedure rather than "a cloud you can trust".
