Vidhi Srivastava

Computer Vision engineer with M.Tech in Artificial Intelligence from IIIT Pune and MCA qualification. Focus on real-time edge perception, model optimization, and production deployment. Expertise in detection, segmentation, tracking, and face recognition. Proficient in deploying optimized models to edge accelerators (Kneron KL630) achieving 25–40 FPS with quantization (FP32→INT4).

Experience with CUDA-based training, transfer learning, and precision optimization.

Computer Vision Engineer at Mantra (Present)

End-to-end C face analytics pipeline development for production deployment. Implemented detection, landmark extraction, quality assessment, and recognition components. Optimized model latency and benchmarked precision/latency trade-offs.

• YOLOv8-Small face detection with reduced latency from 68ms to 35–45ms through quantization
• Sub-center ArcFace + MagFace embedding-based recognition with 1.19% FNMR at 0.1% FMR
• PyTorch AMP mixed-precision training for accelerated GPU training
• Production C deployment on Kneron KL630 NPU achieving real-time inference
• Fingerprint stitching for multi-frame face quality assessment
• Comprehensive precision/latency benchmarking across FP32, FP16, INT8, INT4 formats

M.Tech in Artificial Intelligence – IIIT Pune

MCA