Anuj Dutt.

I architect and ship production AI systems — from cloud-scale GenAI and GPU-optimized LLM inference to edge AI on constrained devices. I specialize in taking models from research to production through hardware–software co-design, aligning models, runtimes, and systems with real constraints in latency, memory, compute, and reliability.

At Adobe, I lead architecture and delivery for GenAI systems across Acrobat initiatives, including PDF Spaces, Contract Understanding and Generate Presentation. My work spans workflow architecture, multi-agent orchestration, evaluation, and production reliability. I set technical direction across engineering, product, design, and research—aligning stakeholders around key decisions and mentoring engineers from experimentation through launch.

Previously, as ML Tech Lead and AI Architect at Jabra, I led edge-AI delivery across three commercial video product lines. I shipped the on-device person detection, tracking and segmentation behind intelligent framing and background effects on PanaCast 50, PanaCast 50 VBS and PanaCast 20 — reducing model memory by 16% and improving performance by more than 20% within a fixed hardware envelope. I owned the SoC platform decision that unblocked a next-generation product architecture, and authored the core algorithm behind the first Intelligent Meeting Spaces release.

Earlier, in Bose's Consumer Electronics Applied Research group, I built on-device AI for the Bose AR ecosystem and delivered Bose's first mobile on-device neural network, shipped in the BoseAR library. I developed an IMU-only system that recognized ten gestures from head motion — without a camera or microphone.