Empowering NYC Tenants to Navigate Housing Laws with LLMs
Written by Andrea Wang
The Challenge
JustFix is a non-profit organization that creates technological tools to empower NYC tenants to exercise their rights to a livable home. They have created many useful tools, such as Letter of Complaint, which sends a USPS Certified Mail on behalf of tenants to the landlords asking for repairs ; or Good Cause NYC, where a tenant can find out if they are covered under Good Cause Eviction law. However, while these tools are very effective and easy to use, they are a small part of the big, mythical maze of NYC housing laws. Navigating NYC housing laws can be very challenging for tenants, especially when facing potential eviction and housing insecurity. Oftentimes, the question is not “am I protected by Good Cause Eviction”, but “I have not heard of Good Cause Eviction. What does it have to do with me?”. That’s why, this summer, I worked with Justfix as a Siegel PiTech PhD Impact Fellow, to find out if they could utilize LLMs to build chatbot-like tools to help the tenants know their rights and what actions to take to assert those rights.
The Project
In the initial, qualitative phase of the project, I conducted site visits and interviews with stakeholders in the NYC housing justice landscape. Specifically, I visited Justfix’s partner non-profit Community Justice Initiative (CJI)’s three physical sites in NYC (called Community Justice Connect), where volunteers are trained to provide legal information for people seeking help on civil legal issues. There, I interviewed lawyers, senior/junior volunteers, and tenants’ rights organizers about their experiences in housing justice. I observed the sites to be more than an information-giving space, but rather, a space of community building and empowerment.
Through these site visits, interviews, and observations, an important takeaway for me was that to build the LLM tool that JustFix is looking for, I needed to navigate an intricate relationship between access to justice and technological barriers. Specifically, many low-/middle-income tenants owning a smartphone do not know its features beyond phone calling, whereas others end up in housing court because they don’t have an email or a required account, or don’t know how or where to scan a document. I also noticed that many volunteers at Community Justice Connect with legal expertise would have to spend time teaching the tenants how to use Whatsapp and Google Translate to understand court documents. This made me realize that any technological tool I would propose would need to address these technological barriers.
I proceeded to design a chatbot tool to mimic how volunteers provide legal information to tenants. For that, I identified that the first thing the LLM needs to do is identify what legal issues apply to the tenants. Effectively, the LLM needs to be able to map tenant narratives about their housing issues to 7 broad areas of housing laws.
Because the technological solution needed to be privacy-preserving and cost-efficient to maintain, I identified that the appropriate models to explore were the local, open-source models such as deepseek-r1, llama, Gemma, and Qwen. To evaluate the models, I collected NYC tenant narratives from public forums such as Facebook and compiled a dataset. The results showed the models to be promising in mapping narratives to legal issues through prompt engineering. For example, deepseek-r1-8b achieved around 70% precision and recall in identifying repair issues, and more than 80% precision and recall in identifying lease, succession, and roommate issues. This meant the models were effective at the initial assessment of what legal issues apply to the tenants.
Impact & Going Forward
To locate a tenant in the housing laws landscape, there are three pieces of information required:
The issues the tenant is currently facing
The tenant’s housing type, and
The tenant’s relevant demographic information, such as disability status.
Yadi Wang
Ph.D. Student, Information Science, Cornell University
While tenants may have different issues from time to time, their housing type and demographic information do not change frequently. Yet, through my interviews with multiple stakeholders, I observed a common pain point – that every lawyer/organizer/volunteer in this housing justice landscape needs to go through the process of finding out the housing type and demographic information. In short, there is no document like a car insurance card or health insurance card that the tenant could just carry when seeking legal assistance. And the process of discovering this information is not structured, but relies on the tacit knowledge of each lawyer/activist/volunteer. Given LLMs’ strength at turning long-form text into structured information, I propose future work that leverages them to design efficient pathways for identifying a tenant’s key demographics and housing type, thereby determing the rights that apply. Through initial explorations, I found that LLMs are quite effective in figuring out these questions through a decision-tree like structure, just with available online resources. By reducing burdens of repetitive tasks and improving access to information, tools like these hold the promise of making housing justice in NYC more responsive, connected, and equitable.