Hack[AI]thon is a 24-hour national-level AI hackathon organized by Sphere Hive, KVG College of Engineering in collaboration with Startup Lab, Joy University.

This hackathon focuses on a problem statement based on Data-Centric AI, where participants improve model performance by improving the dataset rather than modifying the model architecture.

Participants will work on an image classification challenge using the 3LC platform, where they will analyze embeddings, strategically label data, retrain models, and compete on a live leaderboard.

The event is designed to simulate real-world AI workflows, emphasizing data quality, experimentation, and iterative improvement.

Challenge Overview

Participants must build an image classification model using Data-Centric AI techniques.

• Model architecture is fixed
• No pretrained weights allowed
• Performance must be improved by data labeling and curation
• Final evaluation is based on accuracy on a hidden test dataset

Eligibility

• Open to students and developers from all colleges
• Participants can join in teams
• Basic knowledge of Python / Machine Learning is recommended
• Participants must bring their own laptop and charger

Team Format

• Team size: 2 to 4 members
• Cross-college teams are allowed
• Each team must submit one final solution

Rules

• Only the provided dataset must be used
• External datasets are not allowed
• The model architecture must remain unchanged
• Pretrained weights are not permitted
• All work must be done during the hackathon duration
• Submissions must follow the required format
• Any form of plagiarism or unfair means will lead to disqualification
• Judges’ decision will be final

Process

• Participants receive dataset and starter resources
• Train baseline model
• Analyze model behavior using 3LC dashboard (embeddings & metrics)
• Select and label important unlabeled samples
• Retrain model with improved dataset
• Submit predictions to leaderboard
• Final ranking based on accuracy and approach

Submission Format

Participants must submit:

• Prediction file (as per required format)
• Source code
• Brief explanation of approach
• 3LC workflow summary or screenshots
• GitHub repository link (if required)

Evaluation Criteria

• Model accuracy on test dataset
• Data-centric approach and strategy
• Effective use of embeddings and labeling
• Experimentation and iteration
• Clarity of explanation and documentation

Perks & Benefits

• ₹30,000 prize pool
• Certificates for all participants
• Free .xyz domain for participants
• Swag kits and goodies
• Networking with AI mentors and peers
• Hands-on experience with Data-Centric AI tools