Senior Machine Learning Engineer II, NLU & Agentic AI

<h4><strong>The Role</strong></h4> <p>We are looking for an experienced software engineer with machine learning expertise to join us in expanding Moveworks NLU (natural language understanding) and agentic AI capabilities, enabling increasingly magical user experiences and improving Moveworks generative and conversational AI capabilities platform-wide.</p> <p>As a member of the NLU team, you will have all the tools of modern NLP and NLG at your disposal, from best-in-class LLMs, multimodal foundation models, and hybrid vector databases to all the infrastructure needed to fine-tune, evaluate, and serve your own models in production. We are a data-centric team, and you will have the assistance of a <a href="https://www.moveworks.com/insights/moveworks-annotation-philosophy-experts-make-our-ai-better">world-class annotation team</a> to build error-free, inclusive, and privacy-preserving datasets for model training and evaluation. </p> <p>You will also go beyond model training to achieve state-of-the-art AI performance in production, in every meaning of the word “performance”: not just accuracy and quality of outputs, but also latency, reliability, and capability of the end-to-end user-facing system as a whole. Successful machine learning engineers on the team are just as motivated to design and evolve great compound AI systems and their components as they are to train great models.</p> <p>Our team indexes on increasing our ability to move fast, solving challenging product and engineering challenges, and pushing the envelope of value provided to customers. Your work will impact our team’s core objectives to understand every enterprise issue and build the most reliable copilot the world has ever seen, in deep collaboration with other functions within Moveworks.</p> <h4><strong>What You Will Do</strong></h4> <ul> <li>Apply software engineering, machine learning, and <a href="https://bair.berkeley.edu/blog/2024/02/18/compound-ai-systems/">compound AI system engineering</a> to create lasting value for all our customers</li> <li>Take on exciting and difficult challenges in conversational agent domains, such as agent cognitive architecture iteration, multimodal agents, multilingual agents, conversational memory management, reasoning strategies (eg Tree of Thoughts / Graph of Thoughts), <a href="https://www.moveworks.com/insights/moveworks-enterprise-llm-benchmark-evaluates-large-language-models-for-business-applications">fine-tuning LLMs for tool use and enterprise reasoning</a> (including preference alignment with RLHF/RLAIF/DPO), agent evaluation, active learning of exemplars for few-shot text classification, abstractive summarization, and <a href="https://www.moveworks.com/insights/what-is-grounding-ai">grounding verifiability for generated text</a>.</li> <li>Push the envelope of Moveworks commitments to responsible AI, expanding our infrastructure for ensuring models work equally well for all people, red-teaming models to ensure they behave safely and as intended, and keeping our ML at the cutting edge of data privacy and security</li> <li>Use your knowledge of machine learning fundamentals and LLMs to design new algorithms and architectures, evaluate them with small scale experiments and productionize your solutions at scale</li> <li>Research and develop innovative, scalable and dynamic solutions to hard problems</li> <li>Use the latest advances in machine learning and LLMs to enhance our products and create delightful user experiences</li> <li>Spend time weekly reading, discussing, and potentially building models out of the latest ML research and open-source code</li> </ul> <h4><strong>What You Bring To The Table</strong></h4> <ul> <li>Drive to ship product improvements with production-quality, fully unit-tested code and rigorously-evaluated updates to models, prompts, or other tunable system components</li> <li>Ability to solve problems end-to-end with machine learning</li> <li>Solid grasp of model evaluation fundamentals, especia ... (truncated, view full listing at source)