Link to presentation slides: Link
Inspiration
There is a major accessibility problem in finding information online to citizens and government employees.
What it does
Chatify shows all your website's information, accessible in one message. Enter the root URL of any site, and Chatify read and learn the entire site. Chatify-built bots take a question and produce an answer, and are compatible with audio input, output, and 25 languages. For more information, check out our website chatif.ai.
How we built it
Our NextJS front-end is deployed with Vercel. We used the Apify API to recursively scrape entire websites, which then returns a massive flat array of page contents that we convert into a tree structure for more scalable traversal when accessing nodes in the future. Each page in the tree gets a summary generated using GPT-4, then the whole tree is stored in a Firebase Real-Time Database.
When a user asks a question, OpenAI's completion AI finds the node most relevant to the question at each level of the tree by comparing summaries. Once it traverses down to the most relevant page on the site, it gives a specific and tailored answer, all without custom AI training!
Challenges we ran into
Some challenges we ran into include ensuring that our solution had a big, verifiable impact in smart cities.
Accomplishments that we're proud of
We hope that with this tool, we will be able to revolutionize the way that citizens find government information. We were about to give up but persevered.
What we learned
We learned to work as a team of developers, PMs, and designers! It was a really cool and unique experience.
What's next for Chatif.Ai
Check out our figma prototype!
Built With
- apify
- firebase
- next
- openai
- typescript
- vercel
Log in or sign up for Devpost to join the conversation.