Sri Charan Birudaraju

I’m a Vision-Language Research Engineer specializing in foundation models and on-device AI systems, with a focus on practical computer vision: interactive segmentation, multimodal vision applications, and efficient real-time inference.

As Lead Engineer - Computer Vision at Samsung R&D Institute, Bangalore, I led end-to-end development and commercialization of a TinyHQ-SAM-based on-device foundation model for prompt-driven interactive segmentation, including LoRA adaptation, evaluation strategy, and specialized touch and lasso decoders for AI Eraser.

I also built an end-to-end Java–JNI–C++ deployment pipeline to deliver memory-efficient, low-latency on-device inference, and I helped commercialize segmentation for image and motion-photo workflows through the Unified Clipper effort—improving temporal consistency and earning a granted US patent.

My work extends to multimodal segmentation, including a lightweight LISA-inspired vision-language architecture integrating Mobile LLaMA and TinySAM for mask generation, plus video decoding and temporal processing that reduced processing latency by 50%. I bring a data-centric mindset—designing annotation protocols and training strategies—and I’ve reinforced reliability by tackling production challenges, including earlier 4G/5G system engineering and RAN QoS improvements.