Principal Machine Learning Engineer, Applied AI
A senior IC role adapting LILA's scientific AI models to customer workflows through post-training (SFT, RLHF-style methods like DPO/PPO/GRPO), building evaluation loops, and debugging production model behavior. Core stack is Python and PyTorch with deep LLM/multi-modal/agentic AI expertise.
- Compensation
- $252K–$336K / yr
- Location
- Cambridge, MA USA; San Francisco, CA USA
- Work setup
- Not specified
- Seniority
- Principal
- Employment
- Full time
- Experience
- 2+ years
- Category
- Ml / AI
- Posted
Skills
Must have
- Minimum 2-3 years of hands-on post-training experience (SFT, RL methods such as DPO/PPO/GRPO) and evaluation system design
- Strong software engineering skills in Python and modern ML frameworks like PyTorch
- Demonstrated ability to debug ambiguous, high-stakes model behavior quickly using data, traces, logs, and qualitative feedback
- Experience leading technical work across research and engineering teams
- Deep familiarity with large language models, multi-modal models, or agentic AI systems
- Clear communication skills for translating customer needs into technical approaches and explaining complex model behavior
Nice to have
- Experience adapting models for customer-facing or production workflows, ideally in scientific, technical, or data-intensive domains
- Experience with RL post-training, such as RLHF, GRPO, or tool-augmented RL
- Experience building evaluation harnesses, model monitoring, or quality dashboards
- Experience training MoE architectures
- A track record of mentoring engineers and being sought out as a go-to technical expert
Full description
Your Impact at LILA
We are growing our Applied AI org and seeking a Principal Machine Learning Engineer, Applied AI with deep expertise in model post-training, evaluation, and production-oriented ML systems. You’ll shape how Lila’s AI models are adapted to customer-specific scientific needs, with a focus on turning frontier model capabilities into reliable workflows that can be evaluated, iterated and used in real customer contexts.
Applied AI sits at the intersection of AI Research, model engineering, and product deployment. The team partners closely with AI Researchers and Software teams to adapt Lila models to customer workflows, improve model quality through post training and driving how harness engineering shapes model behavior inside the product.
This is a high-impact senior IC role for someone who brings deep, hands-on expertise in model post-training, evaluation, and production ML systems, built over many years of solving this exact class of problem.
What You'll Be Building
- Apply deep, hands-on expertise to close the last-mile gap between Lila's model capabilities and customer-specific scientific workflows.
- Lead post-training efforts using approaches such as SFT and RL (DPO, PPO/GRPO) to align model behavior with customer-specific requirements and feedback.
- Build and run evaluation loops that measure model quality, reliability, and customer fit, using patterns you've refined across many prior projects.
- Turn customer learnings, data signals, and evaluation results into concrete model improvement cycles.
- Partner with AI researchers to translate model advances into reliable, usable capabilities.
- Work with Software to integrate model behavior into end-to-end product workflows.
- Debug complex model failures using traces, evaluations, customer context, and scientific feedback.
- Mentor engineers and share reusable patterns for model adaptation, evaluation, and deployment, drawing on lessons learned across a long track record of shipping ML systems.
What You'll Need to Succeed
- Minimum 2-3 years of hands-on post-training experience (SFT, RL methods such as DPO/PPO/GRPO) and evaluation system design, gained from having solved these problems many times before.
- Strong software engineering skills in Python and modern ML frameworks like PyTorch.
- A demonstrated ability to debug ambiguous, high-stakes model behavior quickly, using data, traces, logs, and qualitative feedback - the kind of judgment that only comes from having seen many failure modes before.
- Experience leading technical work across research and engineering teams.