Staff AI/ML Software Engineer, Model Distillation & Fine-Tuning

General Motors • Mountain View, CA

2d • $189,300 - $290,700 • Hybrid

About The Position

General Motors is bringing multimodal AI into the vehicle, and we are looking for a Staff AI/ML Software Engineer to lead the adaptation, fine-tuning, and distillation of foundation models for the automotive edge. You will build models that understand driver intent, conversational context, passenger requests, and the visual state of the cabin. Large, general-purpose vision-language models (VLMs) and LLMs are highly capable, but their size makes them impractical to run on constrained vehicle compute. Slicing them down naively degrades exactly the reasoning and multimodal ability that made them worth deploying. Solving that is the core of this job. You will join Vehicle Applied AI, the team that identifies, validates, and de-risks the AI capabilities that will define our future vehicles. We prove feasibility on representative vehicle hardware and chart a practical path to scale. As an individual contributor technical leader, you will set the architectural direction for our model optimization pipelines. You will take the lead on parameter-efficient fine-tuning, dataset curation for complex human-machine interaction use cases, and teacher-student knowledge distillation. You will connect foundation model research with practical deployment, ensuring your models understand the cabin environment, improve through continuous data loops, and perform reliably after edge quantization. If you are a strong ML practitioner focused on maximizing the "intelligence per parameter" of compact models, this is the role for you.

Requirements

• Bachelor's degree in Computer Science, Machine Learning, Data Science, Mathematics, or equivalent practical experience.
• 8+ years of software engineering or applied ML research experience, including work where you set the technical direction others built against, made the architectural calls on an ML system, and brought other engineers along with you.
• Deep proficiency in PyTorch.
• Hands-on experience fine-tuning large language models or vision-language models, with results you can speak to in detail.
• Practical experience with at least two of: knowledge distillation, parameter-efficient fine-tuning, pruning, or quantization.
• Based in or willing to work hybrid out of Mountain View, CA or Seattle, WA, reporting to the office three days per week at minimum.

Nice To Haves

• Master's degree or Ph.D. in Computer Science, Artificial Intelligence, or a related field.
• Experience shipping a quantized model to a specific hardware target, including working through the accuracy regressions that surfaced along the way.
• Familiarity with the broader training ecosystem (Hugging Face, DeepSpeed, Ray, or Megatron) and experience managing dataset pipelines at scale.
• Domain experience in conversational AI, human-computer interaction, smart spaces, or deploying multimodal models in consumer-facing products.
• Open-source contributions to foundation model tuning libraries, or published research on model compression, distillation, or efficient AI.
• Ability to communicate complex AI training concepts and architectural trade-offs to cross-functional product and engineering teams.

Responsibilities

• Design and build the knowledge distillation pipelines that transfer reasoning, vision, and language capability from foundation models into compact architectures suitable for edge deployment.
• Apply and scale parameter-efficient fine-tuning techniques (LoRA, QLoRA, or similar) to adapt general-purpose models to specific cabin interaction and conversational AI use cases.
• Build and own the reinforcement learning flywheel, implementing human-in-the-loop alignment (RLHF/DPO) and closing the loop between in-cabin data collection and continuous model improvement.
• Curate, evaluate, and synthetically generate the datasets required to teach smaller models to accurately interpret passenger intent and complex visual cues inside the vehicle.
• Implement Quantization-Aware Training or similar techniques, adjusting model architectures and training regimes to prevent accuracy degradation when models are compressed for hardware deployment.
• Establish the evaluation frameworks and benchmarks for fine-tuned models, measuring hallucination rates, domain accuracy, and safety constraints.
• Own our base model strategy: decide which foundation architectures we build on, and make the case for switching when something better arrives.

Benefits

• Bonus Potential: An incentive pay program offers payouts based on company performance, job level, and individual performance.
• Company Vehicle: Upon successful completion of a motor vehicle report review, you will be eligible to participate in a company vehicle evaluation program, through which you will be assigned a General Motors vehicle to drive and evaluate. Note: program participants are required to purchase/lease a qualifying GM vehicle every four years unless one of a limited number of exceptions applies.
• This Job may be eligible for relocation benefits.

What This Job Offers

• Job Type: Full-time
• Career Level: Senior