Staff AI/ML Engineer, Deep Agentic Reasoning Engineer (Lorenz Labs)

Analog Devices•San Jose, CA

11d

About The Position

We’re looking for Deep Agentic Reasoning Engineer (open rank) to design and build multimodal reasoning models (e.g., time-series, audio, video, sensor data) for Edge AI applications. You’ll collaborate with experts in AI, NLP, audio, time series, RL and other domains to advance reasoning capabilities in alignment with Analog Devices business needs. This role offers the opportunity to work on the forefront of AI research, contributing to the development of agentic systems and multimodal reasoning frameworks.

Requirements

6+ years of experience releasing AI/ML products.
PhD in AI-related fields such as NLP, Computer Vision, Speech Processing, or related discipline.
Strong publication record in top-tier venues (e.g., NeurIPS, ICML, ICLR, ACL, NAACL, EMNLP, CVPR).
Proficiency in training, post-training, and evaluating transformer-based models with deep understanding of training dynamics, architecture modifications, and optimization techniques
Strong expertise in supervised fine-tuning, RL (e.g., GRPO), and advanced post-training techniques for reasoning models, including hands-on experience with RL implementation and policy optimization
Experience in implementing and modifying transformer architectures (encoder/decoder variants, attention mechanisms, positional encodings)
Advanced knowledge of techniques including Chain-of-Thought (CoT), agentic tree search methods, continual learning, etc.
Multi-agent orchestration, task decomposition, and experience with both static frameworks (e.g., LangGraph ) and dynamic multi-agent harnesses for complex reasoning workflows

Nice To Haves

Growth Mindset Appetite for continuous learning with demonstrated ability to quickly adapt to new research directions
Resourcefulness in exploring and implementing emerging paradigms such as JEPA training, world models, diffusion-based generation, and agentic harnesses
Interest in frontier applications including edge AI deployment, multi-agent orchestration, deep research automation

Responsibilities

Develop and Evaluate Models: build and evaluate deep learning models for multimodal classification and reasoning tasks.
Full Research Stack: Engage in the full research lifecycle, including stakeholder discussions, data collection, annotation, preprocessing, model training, and rigorous evaluation.
Prototype and Innovate: Prototype models in Python (using PyTorch/TensorFlow) and implement core reasoning improvements, including mid-training and post-training techniques for multimodal models.
Knowledge Distillation: Apply distillation techniques to transfer reasoning capabilities from larger models to smaller, efficient models.
Dissemination: Document experiments and results through internal reports, blog posts, and publications in top-tier conferences and workshops.