AltCredit

Inspiration

• We were inspired by lack of financial resources in terms of small loans, credit scores, and credit based grants for undocumented people in the United States. We built upon a 2020 paper from Cornell University that used cell data to make pseudo-credit scores for people without SSN data. (https://arxiv.org/abs/2002.12616)

What it does

• Altcred uses a users utility bills, available financial information (pay stubs/W2s), and a users social media activity to generate a pseudo-credit score and approve them for available loans.

How we built it

• We trained a pyspark model using available financial data regarding undocumented people along with their scraped social media profiles. This model then uses a random forest machine learning model to match them with available loans. Their login information and sensitive documents are collected by our React / Next.js frontend architecture, then parsed and processed in our pyspark backend API.

Challenges we ran into

• Implement our system in a way that the user's critical data is transmitted/stored securely

Accomplishments that we're proud of

• Utilization of non-traditional data types to estimate the credibility of people who lack traditional documentations
• A system that continuously trains an AI model to refine predictions
• Human-centered interface to maximize the simplicity of user operations
• Smooth user experience with built-in autofill

What we learned

• Financial knowledge about how credit bureaus generate credit scores and how banks review loan applications
• Feasibility of utilizing AI to improve the current solutions by analyzing diverse data sources

What's next for AltCredit

• Formalize the verification of submitted materials
• Train our model on larger/more realistic datasets to improve reliability
• Customize model's parameters to fit individual banks' needs (?)

Built With

• html5
• javascript
• next.js
• node.js
• pyspark
• python
• react
• tailwind
• typescript