Welfair AI

Inspiration

When looking at different tracks and challenges, the social impact stood out because we wanted to build something that helps people in real situations. We started thinking about common social issues people deal with, and then we realized that people often deal with documents that impact their daily life, such as medical bills, eviction notices, or denied benefits. These can negatively harm someone's life if the right choice of actions isn't taken. This inspired us to create something that can guide people who find themselves in tough situations and break through them.

What it does

Welfair AI is an AI-powered crisis navigation tool that helps people understand and act on important real-world documents. Users can upload documents such as bills, notices, denials, etc. Then, the system explains the situation clearly, identifies urgency and upcoming deadlines, suggests next steps for the user, and provides the option to generate drafts such as emails or appeals to help the user.

How we built it

We built Welfair AI as a full-stack web app with a Next.js frontend, a FastAPI backend, SQLite for local data storage, and the OpenAI Responses API for document analysis. The core flow lets a user upload a document, extract structured case details like urgency and deadlines, save or update a case, and generate action-oriented drafts such as emails or call scripts. We focused on making the experience feel polished and practical while keeping the architecture clean and modular.

Challenges we ran into

One of the biggest challenges was making the AI output feel useful without overclaiming certainty, especially for messy or incomplete documents. We also had to handle issues with conflicting code when pushing and pulling on Github from both our code. Another challenge was balancing speed and polish so the product felt like a real application while still being achievable within hackathon scope.

Accomplishments that we're proud of

Some accomplishments that we're proud of are creating a real full-stack product while also creating a system that heavily prioritizes clarity and trust for the user. We also designed outputs that are more structured instead of common AI responses. Moreover, we built something that could realistically help someone in a seriously stressful situation.

What we learned

We learned that structured outputs are what make AI more reliable than typical chatbots. We also learned that the prompts that go into AI also matter, as it affects the accuracy and clarity of results. We also learned that backend validation is important to ensure that outputs are safe and usable before showing them to users. While in development, we made sure that the frontend, backend, and AI logic were modular to make it easier when it came to debugging.

What's next for Welfair AI

In the future, we could connect users to actual sources such as nonprofits or legal aid. We could also add support for those with multiple languages. We could even add more recommendations and tailor the software for people with their own personalized issues.

Built With

• fastapi
• gpt-5.4-mini
• next.js
• node.js
• openai
• openai-responses-api
• pillow
• pydantic
• pypdf
• python
• react
• sqlite
• sqlmodel
• tailwind-css
• typescript
• uvicorn