WiseConnect

Connecting Generations, Empowering Seniors

Comment

Inspiration

Built for the Rocket Innovation Studio Social Media app challenge.

What it does

WiseConnect is a social media platform with fraud detection capabilities. Our platform has an intuitive UI and font sizing options tailored for senior citizens. Every post and comment is scanned against an algorithm which classifies the text as safe or potential fraud.

How we built it

  • We used Next.js (React) and mantinedev UI library for the frontend. For data fetching from the backend, we used react-query which allowed us to cache async data and the user experience better. For managing our forms we used Formik instead of relying on the normal useState React provides. Zustand was used for managing the client global state.
  • We used FastAPI, an API development library in Python, to develop a front-end REST interface. There are various data validation schemes available in the library that are ideal for rapid prototyping. For the database, we used the simplest of all, SQLite 3. The database schema consists of 5 tables, with all columns having appropriate constraints. Tasks such as tagging posts by analyzing them against the AI are done in the background while the API continues to serve the requests.

Challenges we ran into

  • The biggest challenge was gathering sufficient data to train our machine-learning model. Although there are datasets available, the data is either dirty or outdated. To solve this, we cleaned and merged 3 different datasets
  • The next big challenge was to train a model with this skewed data set with 70% of the data not being scam and only 30% being scam
  • Further, the frontend had to communicate with our APIs for which we need to host our app but we solved this by using ngrok
  • This was the first time we used Mantine dev and we had to refer to the documentation a lot to build a good UI
  • Getting the client state and server state in sync at first was a problem, we solved this partially by using a library called react-query which helped us cache requests and only make calls to the backend if necessary
  • A lot of time was spent in designing database models.

Accomplishments that we're proud of

A working website and backend. We gathered data from various sources and created our corpus of fraud texts. The machine learning model has 95% accuracy.

What we learned

We learned and implemented various elements that make a website suitable for the senior age group.

What's next for WiseConnect

  • Scammers can embed text in images and post them instead of writing as a post, this we can use OCR technology to recognize the text and take further actions
  • Based on the user history, we can develop an algorithm to flag posts as fraud or not fraud

Built With

Share this project:

Updates