• 

  Verity main page

Verity — Ensemble Text Classification System

🌟 Inspiration

We've all seen how a single opinion can be wrong, right? It's the same with AI models.
A lone model, no matter how smart, can miss things and make mistakes.

My inspiration for Verity was the idea of teamwork. I wanted to build a system that brought multiple AI models together to get a more accurate and reliable result—just like a group of experts collaborating on a problem.

⚡ What it does

Verity is a text classification system built on consensus.

• It takes a piece of text and feeds it to a group of different AI models.
  
• Each model gives its own opinion on the text.
  
• A central "boss" model weighs those opinions to make the final, most confident decision.
  

It’s smarter than any single model because it learns to rely on each expert's strengths.

🔨 How we built it

1. Picked a few AI models, each good at a different task.
   
2. Built the system to orchestrate them:
   • Feeding the same input to every model
     
   • Collecting their outputs
     
   • Using a special algorithm (the boss model) to combine those outputs into one verdict
     

🚧 Challenges we ran into

• Getting all the different parts to work together smoothly — like trying to get a group of specialists to communicate for the first time.
  
• Preparing the data: to train a system like this, you need a large, clean, and well-organized dataset, which was time-consuming but crucial.

🏆 Accomplishments we're proud of

• Built a system more accurate than any of its individual parts.
  
• Proved that by working together, AI models can overcome individual weaknesses.
  
• Saw the boss model successfully combine expert opinions into one powerful, correct classification.

📚 What we learned

• The power of ensemble learning — using multiple models to get a better result.
  
• The importance of quality data — even the best idea won't succeed without it.
  

🚀 What's next for Verity

• Add more diverse models to the ensemble.
  
• Explore a feature that gives users a confidence score for each classification, so they can see how sure the system is about its decision.
  

Built With

• dash
• gemini
• matplotlib
• numpy
• pandas
• python
• scipy
• tensorflow
• torch
• transformer