Inspiration
Our mission is to develop and deliver a cutting-edge tool that enables users to actively question and verify the information they encounter. We commit to providing a platform that facilitates access to diverse sources and perspectives, equipping each user with the necessary resources to critically evaluate the validity of information, thus promoting a culture of rigorous inquiry and informational accountability.
What it does
The app has the following workflow:
- Start
- The audio from a video or recording is transcribed into text.
- The OpenAI API is used to obtain claims from the transcription.
- The OpenAI API is used to suggest 5 websites where each claim can be verified.
- A search is conducted for each website, and information is extracted from the pages.
- The information from the websites is contrasted with the help of the OpenAI API to determine how accurate the claims are.
- End
How we built it
We developed a backed in Python (using Flask) to send request to a web server. This server expects to receive audio files, and it connects to the OpenAI API to transcribe and fact-check the claims made in the audio. Then, it displays a summary of the claims in a frontend built in react.
Challenges we ran into
The main challenge was designing effective prompts that would accurately produce the information needed. We needed to iterate multiple times until we got the best prompts that would fact check in the most accurate way possible. Another challenge was web scraping, because some sites blocked requests, and because it is a time consuming process that can be optimized in the future.
Accomplishments that we're proud of
Although our mobile app is still in the prototype phase, it already proves to be a viable tool for combating misinformation on social networks. The functionality to verify the accuracy of facts mentioned in recorded audios or uploaded files, classifying them as true, partially true/false, or completely false, shows significant progress towards the eradication of misinformation.
What we learned
On a technical aspect, we learned how to establish connections with LLMs and use them to solve a social problem. We understood how to work with multiple OpenAI API versions and models. In addition, we learned how to structure a solution that involves multiple algorithmic steps to solve a computational problem.
What's next for Fact Mobile
In the future, we can try different models to get less biased conclusions and avoid hallucinations, like the ones provided by GCP. We expect that this application will be used by companies and individuals to become aware of the vast amount of misinformation on the internet and make the best decision with the information that we have in the moment.
Log in or sign up for Devpost to join the conversation.