A plant or object identification app — photograph something, get back what it is — is a well-scoped, visually compelling student project achievable without training a model from scratch.
A realistic build
- Start with a pretrained image classification or embedding model, not a from-scratch architecture.
- Build (or find) a well-curated reference dataset for the category you're identifying, with clean labelled examples.
- Use similarity search against the reference set (via embeddings) rather than a fixed closed-set classifier, so new categories can be added without retraining.
- Show a confidence score and a few alternative matches, not just one guess — this is more honest and more useful when the model is uncertain.
What makes this project stand out
Handling the 'I don't know' case gracefully — recognising low confidence and saying so — is the detail that separates a genuinely useful demo from a toy that always confidently guesses something.
See image embeddings for visual search.
— Pranjul Rathour, GenAI Engineer from Kanpur, India. Open to GenAI roles, hackathon judging, mentorship sessions and guest talks at any campus: pranjulrathour41@gmail.com.