The posting
Role Overview
We are hiring a Founding ML Infrastructure Engineer to own the end-to-end deployment, optimization, and operation of our suits of models in production. This is a core founding role focused on building and operating production-grade LLM systems. You will apply deep knowledge of model internals to deploy, optimize, and run modern LLMs at scale, owning performance end-to-end across latency, throughput, and reliability.
You will design and operate the full ML serving stack from model artifacts to GPU execution, and work closely with Product and ML teams to ensure our models can support high QPS, strict SLAs, and production correctness. This role is ideal for someone who deeply understands how LLMs work internally, but chooses to specialize in making them fast, stable, and production-ready.
About Realm Labs
Realm Labs is an AI trust and security startup. We help enterprises detect, debug, and prevent AI’s misbehaviors in production. We are backed by top VCs and serve some of the most iconic global enterprises.
Key Responsibilities
Own the end-to-end LLM inference stack, including:Model loading and execution GPU utilization and memory efficiency Runtime performance tuning Production deployment and scaling
Design and operate high-performance LLM serving systems using technologies such as:vLLM, TensorRT / TensorRT-LLM, Triton Inference Server, SGLang
Optimize inference across:Latency Throughput (QPS) GPU memory footprint Cost efficiency
Work hands-on with PyTorch and TensorFlow models, including:Model graph understanding Attention mechanisms, KV cache behavior, batching strategies Precision tradeoffs (FP16, BF16, INT8, etc.)
Build and maintain production-grade GPU services:Multi-model serving Autoscaling strategies Fault isolation and graceful degradation
Collaborate with application and platform teams to:Define serving APIs Ensure correctness and safety of outputs Debug production issues end-to-end
Build a reproducible model training and versioning system for customer deployments Establish best practices for:Model versioning Rollouts and rollbacks Performance benchmarking Production validation
Expected Qualifications
5+ years of professional experience in ML infrastructure, systems engineering, or production ML roles. Strong software engineering fundamentals; ability to write robust, maintainable production code. Deep hands-on experience with LLM inference infrastructure, including:PyTorch (required) TensorFlow (working knowledge)
Proven experience with GPU inference optimization, including:TensorRT / TensorRT-LLM vLLM Triton Inference Server SGLang or similar serving runtimes
Strong understanding of LLM internals, such as:Transformer architectures Attention and KV caching Batching, streaming, and token-level generation
Experience running ML systems in production with high traffic and SLAs Comfortable working in Linux-based, cloud production environments
Preferred Qualifications
Experience deploying LLMs on Kubernetes and GPU clusters. Familiarity with CUDA, NCCL, or low-level GPU performance concepts. Experience with:Model sharding and parallelism strategies Multi-GPU inference Streaming inference systems
Knowledge of observability for ML systems (metrics, latency breakdowns, GPU monitoring). Experience working at startups or owning systems with minimal abstraction layers.
Additional Information
This is a founding, high-ownership role with direct impact on core product capabilities. You will be expected to build, run, and own systems end-to-end. The role may include limited on-call responsibilities aligned with production ownership.
Compensation & Benefits
Market aligned compensation and benefits Founding engineer equity (Equity is a significant component of this role and will be discussed) Medical, Dental, Vision, Life insurance, 401-K, In-office lunch etc.
Visa sponsorship: We do sponsor visas! However, we aren't able to successfully sponsor visas for every role and candidate. But if we make you an offer, we will make all reasonable effort to get you a visa, and we retain an immigration lawyer to help with this.
