Edge AI Multimodal Reasoning Engineer

Analog Devices; San Jose (CA); On-site; USD 140,000 - 210,000; Full time

Job summary

Analog Devices, Inc. through Lorenz Labs seeks a Deep Agentic Reasoning Engineer to design and build multimodal reasoning models for Edge AI applications.

You will collaborate with AI, NLP, audio, time-series experts to advance reasoning aligned with business needs. Responsibilities include building and evaluating deep learning models in Python (PyTorch/TensorFlow), engaging the full research lifecycle, prototyping improvements, applying knowledge distillation, and disseminating results through

Qualifications

• PhD in AI-related fields such as NLP, CV, or speech processing.
• Strong publication record in top-tier venues.
• Proven ability to build and evaluate transformer-based models.
• Experience with multimodal data and reasoning techniques.
• Familiarity with CoT reasoning for LLMs and agentic systems.

Responsibilities

• Design and build multimodal reasoning models for Edge AI applications.
• Engage in the full research lifecycle from data collection to evaluation.
• Prototype models in Python (PyTorch/TensorFlow) and apply mid/post-training improvements.
• Apply knowledge distillation to create efficient models.
• Document experiments and share results in internal and external venues.

Skills

Transformer models

Multimodal reasoning

CoT reasoning

Langchain / langgraph

Reinforcement learning

LLM evaluation

Python (PyTorch/TensorFlow)

Education

PhD in AI-related fields

Tools

PyTorch

TensorFlow

Job description

Analog Devices, Inc. through Lorenz Labs seeks a Deep Agentic Reasoning Engineer to design and build multimodal reasoning models for Edge AI applications.

You will collaborate with AI, NLP, audio, time-series experts to advance reasoning aligned with business needs. Responsibilities include building and evaluating deep learning models in Python (PyTorch/TensorFlow), engaging the full research lifecycle, prototyping improvements, applying knowledge distillation, and disseminating results through