Discover more from Legal AI
Exclusive Interview: Unveiling the Future of Law with AI-Powered Legal Tools
A Conversation with a Pioneer in LegalTech
No Matter Where You Go, You Are on Twitter
Despite all the valid criticisms of the platform, Twitter is still a useful tool to keep updated on the latest in Legal AI tools. I have interacted with more than 50 individuals offering various types of AI powered tools either pointed at the legal market or easily adaptable to it. Juriflow created by serial entrepreneur Mirza Baig is one such tool. He was gracious to take about 40 minutes of out his hectic schedule to sit for an interview about his offering.
First off, Mirza is not a lawyer, but like many non-lawyers, has had contact with so many friends and family who have collided with the legal system. He became aware through those interactions of needs not being served in the legal field that he thought he could fill. More to the point, he thought an AI powered tool could fill one of those needs. Juriflow is one such foray into an open need. He noticed what many others have. OpenAI has marketed their tools, including ChatGPT, in part by emphasizing that its LLM can now pass a standard bar exam correctly answering more than 85% of the questions.
Mirza currently lives in Chicago and was a founding engineer at a midsize startup there. Ultimately his goal was to start his own company and Juriflow is one of those first efforts.
Juriflow’s main page makes the bold marketing statement “Fire Your Lawyer.” He received some push back on that stark slogan on Twitter, but it makes the point he wants to convey. That is, that with AI working for people in receipt of contracts, motions, legal opinions or other legal documents, Juriflow enables them to upload those documents and get ready-made summaries of their key points. It also enables users to have a question and answer experience with that document akin to what they might do with an initial lawyer consultation. Juriflow enables users to pose natural language questions, questions devoid of legalese, and have returned meaningful answers from the AI’s interpretation of the legal terminology in that document. Like many such tools using LLMs, it can be tuned to respond in the style of what is best consumed by a high school graduate, or someone with an 8th grade education or a PhD level education and so on.
Mirza has been busy. “I shipped like more than one and a half products a month up until now.” One of his other products went “zero to twenty thousand users in less than two weeks.”
He describes Juriflow as “your personal AI lawyer.” He designed it to make it easer for anyone to have the “best of the best lawyers for a fraction of the price.” I asked him whether he had heard of the recent attempts by the folks behind donotpay.com to have an AI lawyer speak to a non-lawyer litigant in the courtroom. He had heard of it and the subsequent response by the relevant bar associations but was not concerned about his offering.
Juriflow relies on fine-tuning that Mirza did to focus the model’s responses from something general to those that are more consistent with what a lawyer would offer.
One of his examples for use in a twitter blurb was his uploading of the entire case file for the recent case involving actors Johnny Depp and Amber Heard. From that extensive pdf document (hundreds of pdf pages), users can ask questions about the case such as “how long were they married?”, “What were the alleged defamatory statements in the complaint?” and so on. He can also provide methods to tune the responses to the context of the document. If it relates to a criminal proceeding versus a civil case, the response architecture behind the scenes detects that and answers according to that legal framework. He says that contents assures that the responses are not just “generic intelligence” but tuned to the document’s content. Questions about documents involving different legal specialties cause the AI tool to take on “a different role, different tone of voice” for the “different specialties that they are scoped in.” When the tool engages in the question and answer feature, it not only answers questions about the document relying on the documents’ content, but provides page citations enabling the user to confirm the result or look to that page for further information. I can see this being very helpful for lawyers themselves for 100 page or longer reports, legal opinions, articles, etc.
His inspiration to create the tool with those specialties was a nod to the practice of medicine. He reasoned that different medical providers develop specialties and people with particular medical conditions want a specialist. He reasoned that people using his tool would also not want merely an AI lawyer generalist, but an AI lawyer who has a specialty in the area relevant to the content of the uploaded document.
He is already working on AI powered tools for other areas that touch on the law and will be debuting those later this year.
I asked him about whether lawyers should be concerned about tools like Juriflow and others essentially competing with them in the market. His response was borne of personal experience as a developer.
Even more than the potential impact on the law, tools like ChatGPT have been impacting software developers in a dramatic way. He mentioned that as a software engineer he uses AI to help him build many of the things he is working on. And “it has actually increased my productivity. So I think lawyers should not be afraid.” His take is that welcoming AI into the work of lawyers will benefit them by increasing their productivity so they can focus on other things AI is not as well structured to handle - creativity for example.
Mirza has observed that AI tools already write code “with no bugs, clean-cut precise code [that] just runs properly as soon as you copy [and] paste it. It's that good. And obviously as a senior engineer now, I just have to focus more so on design and architecture.”
Like developers he suggests lawyers seek ways to maximize the use of AI as a competitive advantage. Just as lawyers initially were slow to adopt social media as a means of marketing and communicating, those not “leverag[ing] AI” will be replaced those who do in his opinion.
He thinks LSAT prep, bar exam prep and other aspects of legal education are also ripe for disruption through the inclusion of AI. Places like Khan Academy, a general purpose educational platform, have already debuted their AI tutor technology which they tout as a revolution in education. Imagine a tool that tutors law school graduates as they prepare for the bar exam. The underlying architecture would be the same and is already underway in K-12 education and beyond.
Users of Juriflow do not currently need their own Open API key, but Mirza indicates that eventually longer term subscriptions will need that. An API key is a tool that enables companies like OpenAI to both monitor usage and in some cases charge users via a subscription model to use their tool. Most of his advertising thus far has consisted of tweets marketing the product and even there he has only sent out a handful. Those have driven more than 100,000 views and sparked interest in the product from its release. It’s clear from that response that potential clients are motivated to see what AI lawyer like tools can provide.
The vision for the product for lawyers is to boost their productivity in searching and analyzing long documents. As for privacy, the tool does not store any chat exchanges in a database. None of the Q&A information is stored. The uploaded documents are stored in a user controlled database as necessary to enabling the summarization and chatting, but the user’s interaction with those documents is not recorded to preserve privacy. One suggested use of the tool is enable lawyers to upload their own past pleadings and agreements so they can search all of those documents using natural language searches as part of their everyday workflow.
The tool uses AWS and stores documents in what is known as an S3 (Super Simple Storage) bucket. AWS (Amazon Web Services) by default protects buckets with layers of required access permissions and security. All local chat records and other analysis, however, would be stored on the user’s local computer and never passed to the app or AWS Mirza said.
For the time being, his feature roadmap focuses are all bug fixes behind-the-scenes along with tool which will enable users to create their own motions. This tool is aimed primarily at friends or family members of inmates trying to assist them in preparing motions and other documents when they are unable to afford counsel. He also has plans for a WhatsApp bot that would answer user’s legal questions across a wide range of legal issues. WhatsApp is one of the most popular text messaging apps. It came to prominence years ago enabling people to message each other across the world without incurring text messaging costs by using local wifi networks. Facebook (now Meta) purchased WhatsApp years ago and has been adding features like end-to-end encryption and the ability of developers to create bots since its acquisition. The legal bot would be accessible to WhatsApp users and could be interacted with just like a user would send a message to one of their contacts. The legal bot would be an AI lawyer providing those answers within WhatsApp. How that bot would be trained, tested or refined is not something Mirza has worked out just yet.
JuriFlow feels like the first of many Legal AI apps that will not only challenge the practice of law, but also challenge the regulations surrounding the practice of law itself. Mirza is an independent developer demonstrating the power of the current application creation tools which will spread out the ability of such apps to be innovated both within the U.S. and without. I actually agree with the sentiment he shared that lawyers embracing AI tools where appropriate are going to have a competitive advantage over competitors. It is yet to be seen which tools will be best at providing that advantage, but Juriflow and products/services like it are definitely going to challenge the status quo.