HavenAI

Inspiration

Our project was inspired by a coursework research paper we wrote about the Canadian housing crisis. Through surveys we conducted with residents across provinces like Ontario, Alberta, British Columbia, and New Brunswick, we uncovered the devastating human toll of rental inflation. Our research revealed a stark systemic gap: immigrants and newcomers face disproportionately higher threat levels, with eviction rates for immigrant households in Ontario sitting at a staggering 46.0% compared to 37.9% for non-immigrants. Driven by these numbers, we built HavenAI to turn passive legal data into active, trauma-informed protection.

What it does

HavenAI is an AI-driven housing-related crisis navigation portal for newcomers in Canada. Using a User Intake Form that gathers relevant information about the user, such as location, legal/immigration status, the trouble they are facing (eviction notice/trouble with paying rent/temporary shelters), financial and employment status.

This information is then used to populate a personalized Dashboard that shows the % risk of instability, summary of their assessment, identifies risk factors such as low income and high rent, and directs the user to the Crisis Action Plan page.

The crisis action plan shows a checklist of tasks to be completed in the next day, 3 days and week according to their priorities. It also prompts the user to upload their rental and financial documents on the Document Analysis page in order for them to understand their situation better and get an idea about their rights. It scans documents like lease agreements, eviction notices or tax forms and informs the user of its implications in simple terms. It does not give any legal advice whatsoever and instead prompts the user to get professional legal advice.

The crisis action plan also prompts the user check out and call the matched programs shown on the Settlement Services page. This page lists organisations, services and legal clinics according to the user’s location and rental situation, specifying the contact information, services provided and languages they operate in.

Using this contact information, the user can call the suggested place and press “Start Call Recording” on the Call Assistant page to feed the live conversation with the employee to the assistant. As the conversation ensues, the AI assistant translates and interprets sentences in the selected language, suggests relevant things to ask and at the end of the call, explains what they said and needs to be done and save that chat for future reference.

At any time in their crisis navigation, the user can use AI Chat Navigator to get direct answers to any of their concerns or queries. This tool is designed to keep in mind the psychological state of the user in this situation and the need to explain things in simple terms.

How we built it

Using our knowledge and information gathered from further online research, we made a theoretical blueprint of what the app should look like and the features it should have. Using this rough blueprint we wrote the code with help from Gemini 2.5 Flash, which we then used to input in Base44.

Challenges we ran into

At the start, when we chose our topic, we were familiar with the overall issue but realised that we had to dive deeper to find what exactly the current rental-related public service websites lack that newcomers and immigrants in Canada would really benefit from. This was hard to identify but we figured to put ourselves in their shoes and try to navigate the sites as someone in their situation would. We found that, firstly, there were different sites and programs for different provinces since the housing jurisdiction varies by the province in Canada. Even so, with the ones we checked out, the information was dense and all over the place and we could tell that it would be hard to make an actual action plan. Thus, we obtained our vision for the app that would prioritise personalised recommendations, real-time information and a concrete action plan.

Another hustle was perfecting the prototype itself.. During the earlier stages, we tried hosting the solution on Streamlit by connecting the source code stored in Github. However, we encountered several bugs with the website and struggled with integrating APIs to reflect real-time information about the aid clinics. When we tried to debug one problem, we would lose a different feature. Eventually as we researched more about our architecture and backend, we were able to integrate even more features working in real time. Finally, we input the matured code into Base44’s free trial to get a more professional-looking app prototype.

Accomplishments that we're proud of

We are really proud that even in the current regional-war situation, we were able to attempt and complete our solution till the end. Even though some of us team-mates had varying courseloads and schedules, we were able to connect online frequently enough. Some days one of us even had to miss a few meetings because of health issues, but all our inputs were equally important and valuable in this solution. More than that we are incredibly happy to have built a working prototype for the purpose of coming in use for people facing difficult times and given how we were able to accomplish accurate language translation and real localised program suggestions with working links.

What we learned

Through this project, we got to see how AI can transform immensely stressful and unclear tasks to a guided plan. We saw how important the design of a platform for the user’s convenience and clarity. Most importantly, we learned that AI can streamline processes and tasks, but it cannot replace expert human judgement in such decisive situations. AI is best to use for understanding things and increasing efficiency (such as not needing to browse a dozen sites to find relevant information), but actual advice and decision assistance should be reserved for humans.

What's next for HavenAI

We plan for future versions of HavenAI to expand coverage across provinces, integrate directly with housing assistance applications, and introduce multilingual, voice-based support for users with limited literacy or digital access. We also aim to explore predictive models that can identify housing instability earlier, enabling support before a crisis escalates.