SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Innovaccer is seeking a Principal AI Engineer to lead the design, development, and deployment of production-grade AI systems at scale. You will work across the full AI development lifecycle—from experimentation and prototyping through production deployment and optimization—collaborating closely with product, engineering, and data teams to build LLM-based solutions, AI agents, retrieval-augmented generation (RAG) systems, and intelligent automation workflows.
In this role, you will take ideas from concept to prototype to production, building the model layer of real products while maintaining product-level accuracy bars. You'll write production Python and PyTorch code with deep systems understanding, design rigorous experiments with clear hypotheses and ablations, and build evaluation frameworks before models. You'll measure results skeptically, read current research to distinguish durable techniques from hype, and explain complex AI work to non-technical stakeholders including clinicians and operators.
Key responsibilities include: architecting complete model systems spanning small, medium, and large models; making critical decisions on data mixture, objective design, and training run management; leading large-scale distributed training efforts across hundreds of GPUs on models with tens to hundreds of billions of parameters; building and owning data pipelines including deduplication, filtering, decontamination, and normalization; and setting two-to-four-quarter technical direction for the AI team while driving execution.
You should have an MS or PhD in Computer Science, Machine Learning, or related quantitative field (exceptional BS candidates with substantial research or open-source work considered). Required depth includes: first-author publications at top venues (NeurIPS, ICML, ICLR, ACL, EMNLP) or equivalent research credentials; hands-on fine-tuning of open-weight models with understanding of parameter-efficient vs. full fine-tuning; strong Python and PyTorch with familiarity of modern training/serving stacks (HuggingFace, FSDP, DeepSpeed, vLLM, SGLang); multi-GPU and multi-node training experience at scale; shipped production models with post-training experience; and a track record of finishing ambitious projects.
Preferred qualifications include mentoring other engineers, leading post-training on hundred-billion-parameter models across hundreds of GPUs, designing complete shipped product architectures, making go/no-go decisions on model programs, deep knowledge of distributed training failure modes, and a public record (papers, model releases, systems, or verifiable training results).
