Team
Engineers who lived the annotation problem.
Three people. Combined experience across automotive CV, ML infrastructure, and annotation tooling at production scale.
Vikram Anand
CEO and Co-founder
Vikram spent 8 years building perception systems for autonomous vehicles, with the last three leading a dataset pipeline team responsible for annotating over 12 million frames per year. That role gave him a close-up view of how annotation velocity becomes the gating constraint on model iteration speed. He started Annotgrove to fix that constraint.
At Annotgrove, Vikram focuses on product strategy, early customer relationships, and the long-term roadmap for the pre-labeling system.
Priya Nair
CTO and Co-founder
Priya has a background in large-scale ML infrastructure, having spent time building distributed training systems and data preprocessing pipelines before joining the annotation tooling space. She designed Annotgrove's pre-labeling model architecture and the consensus scoring algorithm that powers the review queue prioritization logic.
Her focus is on model accuracy across diverse object categories, inference latency at batch scale, and the data contracts that make the API reliable enough for automated pipelines.
Marcus Webb
Head of Product
Marcus previously built annotation tooling at a medical imaging company where he worked closely with radiologists and CV researchers who were both annotating data and consuming it. That dual perspective informed his approach to the review interface at Annotgrove: the interface has to minimize annotator decision fatigue while also surfacing the information quality-control leads need to catch errors before they contaminate training sets.
He owns the review interface design, annotator experience, and the export pipeline UX.
How we work
Small team, direct feedback, no annotation theater.
We're a small team. Every product decision is traceable to a real friction point a customer reported or we experienced ourselves.
Small on purpose
Three people means every customer interaction reaches the person who built the feature. No support tiers, no ticket queues. You talk to the engineer who wrote the code.
Real-dataset testing only
We test every model update against a held-out set that includes actual customer dataset characteristics: medical grayscale, automotive dashcam, and retail shelf imagery. Benchmark-only numbers mislead.
Customer data is off-limits
We will not use your training data to improve our models. Full stop. Your annotated frames are not a resource we harvest. They're a dataset you own, and you can delete them at any time.
Ship when it works, not when it demos
We delayed the Pro plan launch by six weeks to fix accuracy on partially-occluded objects. A feature that works in a demo but fails on real datasets is a liability, not a feature.
Want to work with us?
We're a small team, but we're growing. If you're obsessed with annotation quality and ML data pipelines, reach out.