In this role, you’ll be at the frontier of generative AI and foundation models
working on building and maintaining a platform that enables the development
and deployment of large language models and other foundation models fine-tuned
over data from engineering applications. Working with a small, tightly-knit
team, you’ll be principally responsible for designing and implementing deep
learning infrastructure, managing large training and inference clusters over
cloud infrastructure, automating model lifecycle processes, and monitoring
performance at training time and evaluating deployed models. You'll tackle
many challenges, such as building infrastructure to finetune open LLMs on non-
traditional data from engineering applications, handling heterogeneous data
modalities to train multimodal foundation models, and scaling training over
multiple compute nodes in large clusters. If you're passionate about
generative AI, enjoy being on the frontier of applied AI, and love tackling
complex engineering challenges, this role offers you the perfect platform to
make a significant impact while building cutting-edge skills alongside some of
the world’s top experts.
Responsibilities
* Develop training tasks for a generative model capable of engineering physical systems using data curated by domain experts.
* Finetune open LLMs such as Llama and Mixtral with engineering data and tasks.
* Develop a training and evaluation pipeline on multi-node cloud platforms.
* Implement augmentation methods and tool calling to integrate quantitative reasoning capability into LLMs needed for engineering tasks.
* Develop training pipeline for surrogate models that can include multimodal foundation models such as those over pointcloud.
* Integrate the AI/ML pipeline with data curation pipeline using specialized engineering software for simulation and design of physical systems.
* Implement best AI/ML practices to ensure proper storage, formatting, and tool interoperability.
Skills
* Proven experience working on generative AI models such as Large Language Models and MLOps tools such as MLFlow, KubeFlow.
* Experience architecting an AI/ML pipeline for training, staging, and deploying LLMs.
* Experience with PyTorch and LLM ecosystem and libraries such as Huggingface libraries, DeepSpeed, and wandb.
* Demonstrated ability to quickly learn and work with new foundation models.
* Experience with version control systems (e.g.: Git), collaborative software development practices, and continuous integration / continuous deployment (CI/CD) systems.
* Proficiency in Python.
* Strong problem-solving skills and ability to work in a fast-paced, unstructured startup environment.
Preferred skills
* Familiarity with diffusion models and other architectures.
* Familiarity with DevOps tools, Docker/Kubernetes, and TerraForm.
Representative projects
* Setting up an MLOps infrastructure to finetune open LLMs on a set of modelica/simulink designs and their performances.
* Developing and integrating a continuous evaluation pipeline that routinely tests the generalization of a fine-tuned LLM and identifies.
* Developing data curation and monitoring framework to monitor the quality of data and the performance of model, and visualize model performance and gaps across different dimensions.
* Build MLOps pipeline to augment LLMs with retrieval over well-curated knowledge base and function calling to external tools during inference.
About the company:
Our goal at P-1 AI is to develop an artificial general engineering
intelligence—and eventually superintelligence—that can help the human species
design physical systems more efficiently and at unprecedented levels of
complexity. Going beyond existing foundation models, our autonomous AI agent
learns from synthetic training data and real-world feedback and reasons over
an internal multi-physics representation of a product design that encompasses
both geometry and function. We are a world-class team of AI researchers,
engineers, and top industry executives backed by some of the best investors in
Silicon Valley and beyond.
