Sciforium is an AI infrastructure company developing next-generation multimodal AI models and a proprietary, high-efficiency serving platform. Backed by multi-million-dollar funding and direct sponsorship from AMD with hands-on support from AMD engineers the team is scaling rapidly to build the full stack powering frontier AI models and real-time applications.

About the Role

As an LLM Training Engineer, you’ll work across the full foundation-model stack: pretraining and scaling, post-training and Reinforcement Learning, sandbox environments for evaluation and agentic learning, and deployment + inference optimization. You’ll build and iterate quickly on research ideas, contribute production-grade infrastructure, and help deliver models that can serve real-world use cases at scale.

What you’ll work on

This role spans multiple tracks - candidates may focus on one or contribute across several. Examples include:

Pretraining & Scaling

Train large byte-native foundation models across massive, heterogeneous corpora
Design stable training recipes and scaling laws for novel architectures
Improve throughput, memory efficiency, and utilization on large GPU clusters
Build and maintain distributed training infrastructure and fault-tolerant pipelines

Post-training & RL

Develop post-training pipelines (SFT, preference optimization, RLHF/RLAIF, RL)
Curate and generate targeted datasets to improve specific model capabilities
Build reward models and evaluation frameworks to drive iterative improvement
Explore inference-time learning and compute techniques to enhance performance

Sandbox Environments & Evaluation

Build scalable sandbox environments for agent evaluation and learning
Create realistic, high-signal automated evals for reasoning, tool use, and safety
Design offline + online environments that support RL-style training at scale
Instrument environments for observability, reproducibility, and iteration speed

Deployment & Inference Optimization

Optimize inference throughput/latency for byte-native architectures
Build high-performance serving pipelines (KV caching, batching, quantization, etc.)
Improve end-to-end model efficiency, cost, and reliability in production
Profile and optimize GPU kernels, runtime bottlenecks, and memory behavior

Ideal candidate credentials

Technical strength

Strong general software engineering skills (writing robust, performant systems)
Experience with training or serving large neural networks (LLMs or similar)
Solid grasp of deep learning fundamentals and modern literature
Comfort working in high-performance environments (GPU, distributed systems, etc.)

Relevant experience (one or more)

Pretraining / large-scale distributed training (FSDP/ZeRO/Megatron-style systems)
Post-training pipelines (SFT, RLHF/RLAIF, preference optimization, eval loops)
Building RL environments, simulators, or agent frameworks
Inference optimization, model compression, quantization, kernel-level profiling
Building large ETL pipelines for internet-scale data ingestion and cleaning
Owning end-to-end production ML systems with monitoring and reliability

Research orientation

Ability to propose and evaluate research ideas quickly
Strong experimental hygiene: ablations, metrics, reproducibility, analysis
Bias toward building — you can turn ideas into working code and results

Education

MS or PhD in Computer Science, Machine Learning, AI, Mathematics, or related field

Benefits include

Medical, dental, and vision insurance
401k plan
Daily lunch, snacks, and beverages
Flexible time off
Competitive salary and equity

Equal opportunity

Sciforium is an equal opportunity employer. All applicants will be considered for employment without attention to race, color, religion, sex, sexual orientation, gender identity, national origin, veteran or disability status.