Inspiration
Immigrant families in the U.S. often underutilize government assistance programs due to language barriers, confusion around eligibility, and administrative hurdles. According to research, about 13.8% of adults in immigrant families avoid non-cash benefits because of uncertainty regarding how it may affect their immigration status, and nearly half report difficulties with enrollment. These challenges leave many immigrant families under-resourced, deepening existing socioeconomic disparities. We were inspired to create a solution that helps bridge this gap and makes essential services more accessible to those who need them most.
What it does
Our application uses AI to assist immigrant families in understanding and applying for government assistance programs. By engaging users in multilingual conversations, the app collects relevant details and provides personalized, step-by-step guidance in the user’s preferred language. It supports a wide range of programs such as SNAP, FAFSA, Medicare, CalFresh, and ESL classes, offering clear instructions and reducing the burden of navigating these systems alone.
How we built it
We built a custom AI-powered assistant, “Roots AI,” capable of having stateful, ongoing conversations that remember users' names, family situations, income, and housing status. This persistent memory allows families to come back at any time and continue where they left off in the application process.
We utilized:
- Whisper for multilingual speech recognition, supporting 99 languages.
- Google Gemini for deep research, information curation, and dynamic response generation.
- Custom tools to research eligibility, match users to assistance programs, and generate personalized guides.
Challenges we ran into
- Learning how to use react and ai technologies
- Working with auth0 and express
- Learning how to utilize react with typescript
Accomplishments that we're proud of
- Proud of executing idea from ideation, design, and implementation
- Very well thought out figma
- Using a lot of new technologies and resources that we were given through sponsorships
What we learned
- The power of stateful AI in real-world applications is immense—remembering past context drastically improves user experience.
- Multilingual support is more than just translation; it requires cultural sensitivity and accurate interpretation.
- Government systems are complex, and simplifying them through AI requires thoughtful design and deep domain knowledge.
- Accessibility and clarity are essential when building tools for underserved communities—simplicity makes a huge difference.
What's next for RootsAI
We're excited about the future of RootsAI and are already planning several next steps to expand its impact:
Partnerships with Nonprofits & Local Governments: We aim to collaborate with community organizations, advocacy groups, and public institutions to integrate RootsAI into outreach efforts, ensuring it reaches the people who need it most.
Expanding Language Support: While Whisper supports 99 languages, we're working on improving dialect-specific understanding and adding culturally appropriate responses to make the experience more human and relatable.
Offline & Low-Bandwidth Functionality: To support families with limited internet access, we plan to implement offline capabilities and optimize for low-bandwidth environments.
Mobile App Deployment: We're developing a mobile version of RootsAI to improve accessibility and convenience for users on the go.
AI Learning & Feedback Loops: We'll continue training our models based on real user interactions (while maintaining privacy) to improve the clarity, accuracy, and empathy of our responses.
Document Assistance & Autofill: Future updates will include document scanning, auto-filling of government forms, and step-by-step assistance with uploading the correct paperwork.
RootsAI is just getting started—our mission is to build a trusted digital guide that empowers immigrant families to confidently navigate and access the resources they deserve.
Built With
- auth0
- docker
- express.js
- fastapi
- gemini
- letta
- mongodb
- node.js
- react
- typescript
- vercel
- whisper
Log in or sign up for Devpost to join the conversation.