Staff AI Engineer, Applied AI — Smart Vision

Arlo • Milpitas, CA

1d

About The Position

About the role As a Staff AI Engineer for Applied AI, you'll be a technical owner of the models behind Arlo's smart features — from computer vision detectors running on-camera to vision-language models that describe what happened, to the retrieval and agent layers that let customers ask questions about their video. You'll pick the right approach for each problem (train, fine-tune, prompt, or retrieve), prove it with solid evals, and take it all the way to production at consumer scale. This is a hands-on applied role: you ship models, not papers.

Requirements

• BS in Computer Science or a related technical field with 8+ years of experience; MS/PhD in ML, CV, or a related field preferred (or equivalent practical experience).
• 8+ years building production ML/AI systems, with a track record of owning models end to end — problem framing, data, training, evaluation, deployment, iteration.
• Strong computer vision depth: detection, classification, segmentation, tracking, video understanding; you've trained and shipped CV models in a real product.
• 3+ years working with LLMs or VLMs in production — multimodal modeling, prompt and context design, fine-tuning, and evaluation.
• Strong Python + PyTorch; solid distributed systems and cloud fundamentals (AWS, Docker).
• Rigorous about evaluation and data quality — you build the benchmark before you build the model.
• Experience with embeddings and vector search at scale; multimodal or video retrieval strongly preferred.
• Experience shipping LLM agents with tool use, and pragmatic judgment about when not to use an agent.
• Working knowledge of inference optimization (serving stacks, quantization, batching, GPU performance) and cost/latency tradeoffs at scale.
• Demonstrated technical leadership without formal authority: influencing roadmaps, mentoring engineers, driving cross-team decisions.
• Bias to ship, comfort with ambiguity, strong written communication.

Nice To Haves

• Edge/on-device inference (quantization-aware training, pruning, NPU/DSP toolchains) and streaming inference.
• Video processing at scale (decoding, frame sampling, FFmpeg-class tooling).
• Audio or sensor-fusion models complementing video; privacy-preserving or on-device personalization.
• Open-source contributions to CV, serving, retrieval, or agent frameworks; consumer IoT or camera/security domain experience.

Responsibilities

• Build, train, and fine-tune computer vision models for detection, classification, tracking, re-identification, and video understanding, and improve them against real-world customer footage — night, weather, motion blur, odd camera angles, edge compute limits.
• Own our video-understanding pipeline built on vision-language models: frame selection and temporal context, prompt and output-schema design, grounding and hallucination control, multi-event reasoning, and quality tuning for captioning and scene description.
• Adapt models to our domain: SFT, LoRA/QLoRA, preference tuning, distillation into small deployable models, and knowing when a 200M-parameter specialist beats a frontier model.
• Own the data and evaluation loop — dataset curation, labeling strategy, hard-negative and failure mining, active learning, benchmark suites, and offline/online metrics that reliably predict customer-perceived quality.
• Own the embedding and retrieval stack behind video search: multimodal/video embeddings, vector index design and tuning, hybrid search and re-ranking, and natural-language queries over a user's library.
• Build agentic experiences on top of the vision stack: tool/function calling, multi-step reasoning over event history, RAG and memory, guardrails, and tracing/observability for agent runs.
• Keep production inference fast and economical — serving stack choice and tuning (vLLM/TensorRT-LLM/Triton), quantization, batching, GPU utilization, and routing between hosted foundation models and self-hosted open models against clear cost and latency targets.
• Partner with product, data, backend, and firmware/edge teams to turn ambiguous product ideas into shipped AI features, with safe rollout (canary, A/B, feature flags) and real production telemetry.
• Raise the engineering bar — architecture and design reviews, MLOps practices, documentation, and mentoring senior and mid-level engineers.

Benefits

• We’re committed to inclusivity and selecting the strongest candidate—no matter their background.
• Even if you don’t meet every listed qualification, we encourage you to apply.
• We’re happy to support growth in areas essential to the role.
• Arlo is proud to be an Equal Opportunity Employer.
• We value inclusion and are committed to inclusive, and harassment-free workplace.
• We prohibit discrimination and harassment based on all legally protected statuses in all hiring and employment.
• We provide reasonable accommodations to applicants and employees with disabilities, who are pregnant or have a related medical condition, or who have sincerely held religious beliefs, observances, and practices.
• Pursuant to applicable state and municipal Fair Chance Laws and Ordinances, the Company will consider for employment qualified applicants with arrest and conviction records.

What This Job Offers

• Job Type: Full-time
• Career Level: Senior