Sateesh Nori

For many New York City tenants, prodding landlords to make repairs in their apartments can be an uphill battle, often involving long wait times and navigating complex legal procedures with limited access to legal assistance.

To address these challenges, housing attorney and New York University School of Law professor Sateesh Nori developed Roxanne AI, an artificial intelligence-powered chatbot designed to provide tenants with actionable legal information about their rights when it comes to repairs and housing conditions.

Roxanne AI, which Nori said is the first chatbot addressing repairs created in New York, was launched in early January as a collaborative project between the NYU law school, the nonprofit Housing Court Answers, and Australian legal tech company Josef Legal.

The chatbot, which uses an AI technology called retrieval-augmented generation, or RAG, is specifically trained to guide tenants through the legal process of getting necessary repairs, from heating and hot water issues to mold and lead paint concerns.

By offering narrowly tailored advice, Roxanne AI seeks to fill a crucial gap in the city's housing legal landscape, where the right to counsel existing in eviction cases does not extend to those involving repairs.

The chatbot has already been used thousands of times, and early feedback suggests that users are finding it easy to use — and that it's accurate.

"We had a housing court judge last week tell us, 'I tried 100 ways to get a bad answer from it, and it never gave me a bad answer,'" Nori told Law360.

Nori and his team designed Roxanne AI to prevent "hallucinations" and to only provide information within the scope of its knowledge base. The project has drawn attention from legal aid organizations, the New York court system, and even the state attorney general's office, all of whom are interested in how AI might help bridge the access to justice gap.

Here, Nori talks about how he came up with the idea for the chatbot, the issues it seeks to address, and its potential in giving ordinary citizens more access to the legal system.

This interview has been edited for length and clarity.

What is Roxanne AI?

The chatbot is called Roxanne AI, and it is geared towards tenants in New York City who need repairs made in their apartment, and it is the first of its kind in New York that caters directly to consumers. So that's a big deal.

We launched it with no funding from any outside sources. It's a group of volunteers from three organizations. One is myself and NYU Law School. [The second one] is a local nonprofit called Housing Court Answers, which has a staff of 15 people and a budget of about $1 million. They answer a hotline every day from tenants who have problems in housing court. They get about 200 calls a day.

The third organization is a company founded in Australia called Josef Legal. They provide no-code solutions to law firms and corporations. Recently, they built a product called Josef Q, which is an AI RAG bot, an LLM that is fed a knowledge base of information and then can retrieve that information and answer questions in natural language.


How did you build Roxanne AI?

Two years ago, I thought it would be great if we could build a chatbot powered by AI that could address concerns that tenants have. So the first question was, what type of problems would we address? We thought about eviction and rent issues, but we decided against that, because if a chatbot makes a mistake, we didn't want to risk somebody getting evicted. The technology is rather new, and we didn't want to sink the entire enterprise by engaging in something that's risky. So we picked the topic of repairs and this is important for a number of reasons.

Number one, it's very hard for tenants to get help when they're facing repairs or poor conditions in New York City. Although there is a right to counsel for tenants who are facing eviction, that right doesn't apply in cases in which the tenants are fighting for basic needs like heat and hot water, working elevators, fighting against lead paint and mold. If tenants are to seek help, they either have to call a hotline like Housing Court Answers and wait 45 minutes to an hour to speak to somebody; or they have to travel to a courthouse or legal services office and sometimes wait for three to four hours before they are seen, only to be turned away in most cases because they're not facing eviction.

So we introduced the topic of repairs. I matched up Josef Legal with Housing Court Answers. We built a team of about five people, and that team worked for 13 months to build a knowledge base for all the information we could find about repairs in New York City — we included articles, statutes, training, materials, know-your-rights guides and so on — to be fed into Josef Q and then get vectorized, which is the way that large language models process data.

And then, we built a little model that we tested thousands of times. We asked many, many questions, and there were a number of goals. Number one, we didn't want the bot to hallucinate. So if the question from a user is outside the scope of repairs, we trained the bot to say, "I don't know."

That's a really big feature of the bot. We don't want the bot to venture outside of its expertise.

Second, we tried to ask questions in as many ways as possible, so that the bot would pick up the intention, the nuance, the varying use of language across New York City. So you could type in "no heat" or "I don't have heat" or "I need my heat" or "the landlord isn't fixing my heat."

All of those questions will result in a detailed list of action items.

What has the response been to Roxanne AI?

Over the last three to four weeks, the bot has been used at least 1,000 times. A lot of people are asking about heat. January has been particularly cold. If you have young kids or elderly relatives and they don't have heat, that can be life-threatening.

Other things that we've heard are that it's easier to talk to a bot because sharing information about your apartment conditions can be embarrassing to people. We also got a comment from somebody who is neurodivergent and said it's easier to communicate by typing than to listen to someone on a hotline or to read a guide that's on a website.

Obviously, we have to do a lot more user feedback — but let me say, we don't have any funding.

We are building this with a bunch of volunteers. Everyone has other jobs, and so I'm just really proud that we were able to launch it at all, that it's in the public now and anyone can use it. AI is shrouded in mystery for so many, especially lawyers. Our team was eager to bring Roxanne AI to light in order to prove what is possible and to push the boundaries of what we can imagine.

For us, it's not even about the AI, it's about the results. AI is the tool that we use, but the results are getting people to attend to their needs, getting their apartment fixed, getting their elevators repaired. And maybe, at some point, they'll need a lawyer, but in the beginning, there are many things they can do. That's the key.

We're in the first month, and we've gotten a lot of positive feedback. We have already demo'd the tool, at their request, to a working group composed of representatives of the court system, the New York attorney general's office, and large nonprofits. There's a lot of excitement about replicating this type of thing in other contexts, which I think is great.

A lot of people still don't know that this exists, so we're trying to get the word out.

What kind of positive impact can a platform like Roxanne AI have on access to justice?

I think there's so much need in New York City and everywhere else for people to get the information that they need to solve their basic problems, whatever that might be, before they become serious legal problems.

Roxanne AI is going to free [attorneys] from the burden of intake and triage, where all the problems are equal until they're actually assessed by human beings. Essentially it will allow the lawyers and the professionals to focus on the hard problems and leave the easy problems to AI.

The easy things are, for example: Where is the courthouse? What type of form do I fill up for this type of problem? What is the nearest place where I can ask for help?

We can train the AI to answer these questions an infinite number of times.

The other thing I'm looking forward to is having Roxanne AI available in Spanish. So many Spanish speakers face language-access issues when they go in person to meet with somebody, or they have to use an interpreter on a hotline, or they have to use a court interpreter who may not be as proficient on technical legal terms. Well, with AI, we can solve these problems. And if "Roxanne Espanol" comes out, or "Roxanne Bengali," think of all the people we could reach that are invisible right now or unable to get help in the current system.

How does Roxanne AI work, in a nutshell?

If you compare an AI tool like this to a Google search, you'll see many, many differences. A Google search will just give you a bunch of links, some are accurate, some are inaccurate. You'll get a bunch of ads. You'll get information from other places outside of New York City.

But if you use Roxanne AI, it will tell you, within five or six steps, exactly what you need to do and in what order for your particular problem. That's huge. That is the type of help that before AI you could only get from a legal professional: narrowly tailored, actionable legal information.

I'm really excited about this. I've been invited to conferences all over America to talk about Roxanne AI and to present a demo of it to other organizations that do legal services work, to talk to law schools, to talk to court systems, to talk to funders. There's a lot of excitement. And what's really amazing is that this tech is cheap, easy to build, and it can work in any context. You just need to feed a specific knowledge base to the bot.

Court systems remain complex to navigate for many people. AI helps litigants access those systems, but could technology do anything to change and simplify the very way courts operate?

I have two thoughts on that. One is that the reason the courts are sometimes viewed as bad by the average person is that the complexity favors professional actors — people who go there over and over again like lawyers, law firms and the government.

Courts are just like a giant game with a set of defined rules. AI is really good at rules. If we can make it easier for more people to go to court, we can try to change that dynamic and level the playing field and make the courts change and address everyday needs of people. We can have just as many ordinary people raising their claims as corporations can.

The second thing is that there's a lot of hope both from the outside — we can get more regular people into the courts using this technology — but also from the inside: The courts can also create new pathways to navigate these complexities for ordinary people.

The other thing about AI is that, if we build a bot, it's available 24 hours a day, seven days a week, 365 days a year, and we can make it work in almost any language spoken on Earth.

What person can do that? How many people would it take to meet that kind of demand?

So, I think all of these things will take some time, but it will really open up pathways to justice for people.

How can you ensure that a bot like Roxanne AI is accessible to all people, including those who are not educated to use digital platforms?

The bot works on a smartphone, on your tiny screen. That's a big deal, because most people don't have a laptop or broadband internet. This type of technology has to be available on mobile devices.

The second way that this technology is helpful is that it can help [most] people in advance so that they don't have to come to a building and wait in line. That results in more capacity for the human workers to help the most vulnerable people — the person who might be speaking a language that's not available, or who has no access to technology, or the person who's extremely elderly and they need a human being to explain things and help them very slowly.

Right now, we don't have time, because everybody's on the same line, and they're equally important until we talk to them and see what the problem is.

What does a project like Roxanne AI mean to the overall goal of increasing access to justice?

I think this project is going to help a lot of tenants in New York City, but even more importantly than that, it serves as an example of what's possible.

If people in other situations — for example in the immigration context, in the bankruptcy context, the eviction context — are able to see that this is possible and build similar tools, that would be the biggest impact Roxanne AI would have.

The goal is to bridge that gap, where most people cannot get a lawyer to help them, whether it's for free or paid. There just aren't enough lawyers available. People don't have that access, and that impacts their lives in numerous ways.

--Editing by Orlando Lorenzo.

Have a story idea for Access to Justice? Reach us at accesstojustice@law360.com.