UK-founded enterprise legal technology provider Panoram has launched a multi-model GenAI document review and data analysis platform that it says will transform, not just speed up, document review at scale, such eDiscovery, due diligence, and cyber breach processes.
PanoramAI utilises a range of GenAI models interchangeably, dependent on the type of use case, which it refers to as an AI matrix. In developing the platform, Panoram looked at how multiple models perform differently, and established that the use of single models can reduce effectiveness but also be more expensive for the user.
The platform currently has a dozen large language models that the Panoram team has analysed and tested on data sets and given a comparative F Score – a measure of predictive performance calculated from the precision and recall of the test. Customers are now able to come to the PanoramAI Lab to test the models they want to use for particular use cases.
Panoram was co-founded in 2020 by former senior Neota Logic executives Rick Seabrook and Greg Wildisen and has since grown to 30 people. The company at its core has focused on eDiscovery work and speaking to Legal IT Insider, Wildisen said: “We’ve been spending a lot of time building a platform that is LLM agnostic in the sense that we use multiple models. The power is in replacing humans for document review. We’re seeing a lot of AI agents and bots unveiled and the big platform players are adding AI capabilities to make lawyers an hour-a-day more efficient, but we as a business see there being much more transformative power in things like document review: it’s not about making it more efficient, but replacing document review completely with GenAI technology.”
PanoramAI was recently put to the test on a large-scale document review exercise following a cyber breach response, where traditional methods of review had been used. The platform reviewed documents at a rate of 7,500 documents per hour, while the rate of human review averages about 40 documents per hour per reviewer. The potential cost differential is £250k versus £20k.
Wildisen said: “The bigger the project, the more valuable the technology is.” He added: “We had to extract the PII but the data that was most important to the customer was the passwords – if you think about the complexity of finding password in electronic data, you’re basically looking for anything but certain LLMs – not all of them – are exceptionally good at finding that kind of complex information and much better than humans.”
Another thing that PanoramAI is good at is OCR and Wildisen said: “We now know which models and combination of models give the absolute best results and and it’s not only in things like relevancy, review and extraction, but it’s also in some of the other areas where technology has, I suppose, hit a bit of a glass ceiling. One of the one of the reasons why discovery and review is difficult is where you have things like a certificate of title, an old document. If you run that through an ordinary discovery tool, this is what you get (see image immediately below). This is the quality of the OCR you get and you can see the text you get is rubbish. With PanoramAI, it even understands that there’s a signature and tells you there’s a signature.” (see second image below)
The human F-Score for the cyber breach exercise was 0.7 compared with the LLM F-Score of 0.9.
Wildisen adds: “What’s interesting is that often with new technology its either faster or better or cheaper, but it’s rare you get all three. I think the exciting thing about this is that its faster, better and cheaper, and when you get all of those three together that’s what creates real disruption.”
PanoramAI is zero data retention as the platform prevents models being trained on client data and it is ISO 27001 certified. You can find out more at https://panoramdigital.com/.