Applied AI Engineer

Norbert Health•Brooklyn, NY

70d•Hybrid

About The Position

Norbert is developing autonomous robots for healthcare delivery. Their AI sensing platform enhances existing robotic systems to assist care teams by performing patient rounds, contactless vital sign monitoring (FDA-cleared for pulse and respiratory rate), running assessments, documenting in the EMR, and escalating critical situations. The company is addressing the global nursing shortage, with deployed systems currently monitoring hundreds of patients daily. Norbert is a small, international team backed by prominent VCs, with offices in Brooklyn, Paris, and Montreal. This role focuses on transforming manually run ML workflows into automated, production-grade MLOps pipelines for reliable deployment on robots in nursing facilities. The ideal candidate will have deep knowledge of the model landscape, prioritize evaluation as a key engineering task, and possess strong opinions on model integration strategies (prompting, RAG, fine-tuning, etc.). The role involves working across cloud and edge deployments, including systems on a SaMD pathway, requiring comfort with regulatory constraints.

Requirements

BS in Computer Science, Engineering, or a related field, or equivalent hands-on experience.
4+ years shipping ML/AI systems in production outside of academic settings.
Strong working knowledge of the modern foundation model landscape (open-weight LLMs and VLMs, common detection/segmentation backbones, embedding models).
Hands-on experience with PEFT/LoRA and supervised fine-tuning.
Strong Python skills; comfortable with the deployment toolchain (ONNX, quantization, at least one inference runtime—TensorRT, vLLM, llama.cpp, etc.).
Experience with a cloud ML training/MLOps platform (GCP Vertex AI, AWS SageMaker, Azure ML, or equivalent).
Ability to work independently, solve complex problems, and drive projects to completion.

Nice To Haves

Edge ML deployment (Jetson, ARM, mobile NPUs).
Real-time voice AI pipelines (STT, TTS, streaming LLM).
Production RAG systems beyond toy implementations.
Experience with medical devices, SaMD, or other regulated ML environments.
Familiarity with MLOps tooling (Weights & Biases, MLflow, DVC, etc.).
Experience with active learning or human-in-the-loop labeling workflows.
C++ for integration with the computer vision pipeline.

Responsibilities

Integrate foundation models and ML components (VLMs, LLMs, ASR/TTS, detection/segmentation, embeddings) into production pipelines, utilizing both open-weight models and third-party APIs.
Build RAG and agent-style orchestration for clinical reporting and conversational interfaces.
Ship real-time streaming pipelines (voice agents) alongside batch and request-response workloads.
Build evaluation harnesses to detect regressions across model swaps and measure performance against clinical-grade accuracy targets.
Fine-tune and retrain models (LoRA, PEFT, supervised fine-tuning) using data collected from the deployed fleet.
Deploy across inference surfaces: third-party APIs, self-hosted, and on-robot edge.
Build the data flywheel: pipelines for collecting, labeling, versioning, and feeding production data back into model improvement.
Partner with the algorithms team (signal processing, computer vision) on integration with their lower-level pipelines.