AI models are only as good as their training data, but validating datasets is time-consuming.
Create a web app that scans datasets for anomalies, biases and errors before training.
AI startups and research labs would pay for this to prevent wasted training cycles.
Start with basic statistical analysis of common dataset formats.
The risk is established players like HuggingFace adding this as a feature.