💡 About the Challenge

Welcome to the State-of-the-Art AI (SOTA-AI) Monthly Challenge.

In an industry flooded with API wrappers and prompt engineers, we are searching for the true builders. SOTA-AI is a student-driven initiative founded by medalists from the International Olympiad in Artificial Intelligence (IOAI). We operate on a simple belief: true AI mastery requires versatility—the ability to handle structured data, natural language, and visual information with equal rigour.

This is not a hackathon where you stitch together pre-made services. This is a Kaggle-style engineering marathon designed to test your depth, not just your speed.

To conquer this challenge, you must tackle three distinct tasks. We are testing your ability to generalize your skills across the entire AI landscape.

⚠️ Important Note: The exact problem statements, datasets, and evaluation metrics for the tasks below will be revealed on Day 1 of the challenge, both here on Devpost and on our Discord server.

📊 Task 1: Classical Machine Learning

• The Domain: Structured / Tabular Data.

• The Challenge: It all starts here. You will face a raw dataset requiring extensive feature engineering, handling of missing values, and the application of statistical learning models. This is about mathematical intuition, not just importing libraries.

• The Goal: Prove that you understand the fundamentals—Regression, Classification, and the bias-variance tradeoff—before you attempt deep learning.

💬 Task 2: Natural Language Processing (NLP)

• The Domain: Text & Sequence Data.

• The Challenge: Move beyond simple API calls. You will be tasked with building, fine-tuning, or optimizing models that understand context and semantics. Whether the task involves sentiment analysis or sequence generation, you must own the architecture.

• The Goal: Demonstrate your technical understanding of Embeddings, RNNs, or Transformer attention mechanisms.

👁️ Task 3: Computer Vision (CV)

• The Domain: Image & Pixel Data.

• The Challenge: Teach a machine to "see." You will work with image datasets to perform classification, segmentation, or object detection. You will need to handle data augmentation and architecture selection.

• The Goal: Showcase your grasp of Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs).
  
  

• Problem-specific metrics