ERCDestroyer

How to run it

Frontend

Enter the tiktok clone directory create a .env in the tiktok directory insert supabase URL and ANON key

npm i 
npm start

Video Analyser

create .env file under video_analysis directory insert your google api key install necessary library run it!!

Inspiration

Many creators face low visibility on TikTok and often feel discouraged after posting several videos that receive few views. We noticed that the current value-sharing system relies heavily on simple creator metrics, such as views and likes, which may not fully capture the value a creator generates.

We were inspired to enhance the value-sharing system by using AI to analyse and measure the value of content more holistically, ensuring that creators are rewarded fairly for the value they contribute.

What it does

• Web Application: A frontend dashboard that displays analytics and insights about a creator’s videos. The dashboard improves transparency in the value-sharing system by providing detailed information on how revenue is distributed and the sources of earnings.

• AI Content Scoring System: An AI model to evaluate videos for marketability, monetisability, and audience engagement.

• Fraud Detection ML Model: A machine learning system to detect fraudulent transactions in the reward system.

How we built it

1. Frontend:

   • Built with React and TypeScript.
   • Displays analytics, content insights, and engagement metrics in an intuitive dashboard.
   • Improve upon the

2. AI Content Scoring:

   • Integrated a LLM for multidimensional content scoring using Google Gemini.

3. Fraud Detection:

   • Implemented an XGBoost classifier to detect suspicious transactions.
   • Features used include transaction patterns, timestamps, and user behavioral metrics.

Challenges we ran into

• Optimising the LLM to process inputs and generate outputs efficiently.
• Designing prompts to make the LLM produce accurate, relevant results.
• Handling high false positives in the ML fraud detection model.
• Dealing with inconsistent data sources during model training.
• Optimising the ML model to efficiently process large volumes of transaction data.

Accomplishments We’re Proud Of

• Developed a content scoring system that goes beyond simple engagement metrics to evaluate the true value of a creator’s work.
• Achieved high accuracy with the fraud detection system, effectively identifying suspicious transactions.
• Built a dashboard that allows creators to easily view insights and feel proud of their content and achievements.

What we learned

• Technical: Learned how to integrate LLMs with web applications, use XGBoost for classification, and process high-dimensional video features.

• Product Insight: Understood the importance of multidimensional metrics beyond simple likes and views for creator value assessment.

• Teamwork & Problem-Solving: Coordinated AI and frontend development and overcame challenges in integrating models with the live dashboard.

What's next for ERCDestroyer

• Improve the efficiency and speed of the content scoring system.
• Explore alternative ways to incentivize creators beyond monetary rewards.
• Implement a more robust peer-to-peer payment system.
• Develop an algorithm to determine whether a creator should receive a flat revenue rate or a higher revenue split, encouraging smaller creators to continue producing content.