Hai Tran

Ninja Van

Designs and deploys computer vision models for package sorting, damage detection, and warehouse inventory tracking. Owns the full MLOps pipeline from data collection to production inference.

• CV models process 500K+ package images/day with 98.2% classification accuracy
• Damage detection system reduced manual inspection costs by 60%
• Inference latency under 85ms on GPU-backed endpoints via TensorRT optimization

Lazada

Built visual search and product recommendation models serving millions of users. Implemented A/B testing infrastructure for model deployment.

• Visual search feature increased product discovery by 28% for 8M monthly users
• Recommendation model lifted conversion rate by 15% across 3 product categories
• MLOps pipeline reduced model deployment time from 2 weeks to 4 hours