A tech start-up recently announced that it has been granted seven U.S. patents for a system that applies a “deep learning” algorithm to examine corporate e-mail databases and flag those with message fields or attachments containing language that might increase risk for a company involved in a federal discrimination lawsuit.

The company says its goal is to help businesses avoid litigation by providing corporate counsel with early warning of litigation risk in their documentation troves, so the sources of potential trouble and questionable activities can be addressed before they become fodder for legal opponents.

“Out of just over 20,000 e-mails, it can cut it down to 0.25% for you to look at,” says Nick Brestoff, founder and CEO of Intraspexion and the inventor listed on all seven patents, although one medically-oriented patent was co-invented with his son, a doctor who envisions taking a similar approach to examining large volumes of electronic health records to correlate patient symptom complaints with diseases.

In e-mail exchanges, Brestoff says he was a California litigator for 38 years. Before law school, though, he says he received a BS in engineering systems from UCLA and an MS in environmental engineering science from Caltech. After retiring, he says he founded his company, Deep Learning, “and started my second childhood.”

Brestoff says in the initial application, the Deep Learning engine has been trained on categorized text from settled lawsuits. The engine uses the language that supported claims to “pattern match” that text with text contained in vast troves of emails within corporate databases.

Only the "related" emails and their scores are reported out for human review. “So the engine isn't substituting for a human; it's augmenting the reviewers intelligence by fishing up something a reviewer would care about from the vast data lake of digitized words,” Brestoff says.

Another potential application of the tool is product liability, which is one way the system could apply to construction, the inventor says. “When things go wrong and there's a failure [of a product], forensic engineers are called in to figure what happened and why.

After-the-fact it often is discovered that, internally, there was a conversation about a risk and the need for a recall, but ‘push back’ from a manager concerned about cost control quelled it. “But forensic engineers are also interested in prevention, and that's where Intraspexion may apply,” Brestoff adds,

“Deep Learning is a multi-layered neural network that also learns from mistakes and back-propagates to minimize errors, so, perhaps in the construction industry, there may be many text-based examples of warnings in post-incident reports.” The trigger language in such reports of settled incidents could be used to train a construction-defect pattern-matching algorithm to search company files for hidden trails that may help spot, and correct defects in the future.

“There may be a thousand post-incident reports concerning some similar category. Could any engineer working on similar project remember them? No. But a Deep Learning model of that risk can "pattern match" for it. It augments your intelligence so you can go from managing a lawsuit to investigating the risk of one, and go from being reactive, to being proactive.”

Although the initial focus is on employment discrimination, the company’s patent applications suggest many other purposes are anticipated. The patents contain 120 claims and cover implementations of Deep Learning to identify general risk, medical risk, entertainment risk, contract drafting risks, product defects, and one patent to support the identification of financial advantages like R & D tax credits.

Intraspexion offers to do a limited number of free pilot projects with corporate legal departments that have recently settled employment-discrimination lawsuits. Brestoff adds, Intraspexion is a "pre-revenue" startup, but "it has a functional system" and is in a pilot now with an NYSE-listed transportation conglomerate, whose name he would not disclose. "If you know of a construction company that's looking at Deep Learning now, I'd love to work with one,” he says.