Image Moderation and Computer Vision Intern

Why this role exists

Recruiters upload company logos, event photos and scanned certificates, and none of it is checked before it goes live. Across three months you will build the visual quality layer that sits in front of that upload flow.

The build

Your first milestone is a lightweight PyTorch classifier that separates usable logos from blurry crops, screenshots of screenshots, and images with text burned across the middle. After that you move to OCR, pulling text out of uploaded certificates so the moderation queue can flag mismatched names automatically. You assemble the training set yourself from roughly eight thousand stored uploads, hand label a validation slice, and argue for whichever augmentation strategy survives contact with real data.

Mentorship and terms

You sit with two engineers in the Pune office three days a week and demo to the platform team every fortnight. Code review here is thorough and kind, and you will be asked to explain your evaluation metric before anyone looks at your accuracy number. This position is unpaid, and the trade is a documented, shippable project plus the review habits that come with working alongside people who care about them. Final year students and recent graduates are welcome to apply.