Deskripsi Pekerjaan
At Nexus Intelligence Labs, we are architecting the future of generative AI infrastructure. We aren't just fine-tuning models; we are building the high-throughput, low-latency serving engines that power the next generation of autonomous agents.
We are looking for a Senior Machine Learning Engineer who thrives at the intersection of deep learning research and high-performance distributed systems. You will own the lifecycle of our core model serving stack—from kernel-level GPU optimization to designing scalable MLOps pipelines that deploy updates to millions of users with zero downtime.
This is a high-autonomy role for a builder who wants to see their code directly accelerate the intelligence ceiling of our platform. You will work alongside researchers from top labs (FAIR, DeepMind, OpenAI alumni) and infrastructure engineers who have scaled systems at hyperscaler magnitude.
Why This Role is Different
- Research Velocity: Push model updates from experiment to production in hours, not weeks.
- Hardware Access: Dedicated H100/A100 clusters; no queueing for compute.
- Ownership: End-to-end ownership of the inference stack—no handoffs to separate platform teams.
- Impact: Your optimizations directly reduce COGS and unlock new product capabilities.
Tanggung Jawab
- Design, implement, and optimize high-throughput, low-latency model serving infrastructure (TensorRT-LLM, vLLM, custom CUDA kernels) for LLMs and multimodal models.
- Architect and maintain automated MLOps pipelines (CI/CD/CT) for training, evaluation, quantization, and deployment across hybrid cloud/on-prem GPU clusters.
- Collaborate with Research Scientists to productionize novel architectures (MoE, RWKV, State Space Models), bridging the gap between prototype code and hardened production artifacts.
- Lead performance profiling and bottleneck analysis (NCU, Nsight, PyTorch Profiler) to maximize MFU (Model FLOPs Utilization) and minimize TTFT (Time To First Token).
- Build robust observability frameworks for model drift, data quality, and system health (latency percentiles, error rates, GPU utilization).
- Mentor junior engineers and establish best practices for GPU programming, distributed training (FSDP, DeepSpeed), and inference optimization.
- Contribute to open-source inference ecosystems and publish technical blog posts representing Nexus engineering brand.
Kualifikasi
- 5+ years professional software engineering experience with 3+ years focused on ML Infrastructure / MLOps / Inference Systems.
- Expert-level proficiency in Python and C++; deep understanding of memory management, concurrency, and SIMD/GPU programming models.
- Proven track record optimizing LLM Inference (speculative decoding, continuous batching, PagedAttention, quantization: GPTQ/AWQ/FP8).
- Hands-on experience with Kubernetes (operators, schedulers for gang scheduling), KubeFlow or Ray, and infrastructure-as-code (Terraform/Pulumi).
- Deep familiarity with PyTorch internals (TorchScript, TorchDynamo/Inductor, custom autograd functions) and Triton Lang / CUDA kernel writing.
- Strong grasp of distributed systems concepts: consensus, sharding, load shedding, backpressure, and observability (OpenTelemetry, Prometheus, Grafana).
- Bachelor's or Master's in Computer Science, Electrical Engineering, or equivalent practical experience; PhD a plus but not required.
- Ability to thrive in a fast-paced, ambiguous environment; bias for action and pragmatic engineering over academic purity.