Deskripsi Pekerjaan
Shape the Future of Generative AI
At Apex Cognitive Systems, we aren't just building models; we are architecting the cognitive layer for the next generation of enterprise software. Backed by top-tier VCs (Sequoia, a16z) and trusted by Fortune 500 clients, we operate at the intersection of foundational research and massive-scale production.
We are seeking a Senior Machine Learning Engineer to lead the lifecycle of our core LLMs and multimodal systems. You will own the path from data curation and distributed training to low-latency inference optimization. This is a high-autonomy role for an engineer who treats model deployment as a first-class software engineering discipline.
Why This Role Is Different
- Compute Access: Dedicated access to 1000+ H100 GPUs; no queueing for resources.
- Open Culture: We publish at NeurIPS/ICML and open-source core tooling.
- Impact: Your models power decision-making for billions in commerce daily.
Tanggung Jawab
- Architect & Train: Design and implement large-scale distributed training pipelines (FSDP/DeepSpeed) for 7B–70B+ parameter models on multi-node GPU clusters.
- Optimize Inference: Deploy models using TensorRT-LLM, vLLM, or custom kernels to achieve sub-100ms latency at high throughput; optimize KV cache management and quantization (GPTQ/AWQ/FP8).
- Data Flywheels: Build automated curation, filtering, and synthetic data generation pipelines to continuously improve pre-training and RLHF datasets.
- Evaluation Rigor: Develop robust offline/online evaluation frameworks (human-in-the-loop, LLM-as-a-judge, red-teaming) to measure reasoning, hallucination rates, and safety alignment.
- MLOps Leadership: Own the CI/CD/CT (Continuous Training) infrastructure: experiment tracking (Weights & Biases/MLflow), model registry, feature stores, and automated rollback strategies.
- Cross-Functional Collaboration: Partner with Product and Research to translate ambiguous research ideas into production-grade APIs; mentor junior engineers on GPU optimization and system design.
Kualifikasi
- Education: MS/PhD in Computer Science, Machine Learning, or equivalent practical experience (5+ years industry ML experience).
- Deep Framework Expertise: Expert-level PyTorch/JAX internals; proven experience writing custom CUDA/Triton kernels or optimizing transformer attention mechanisms.
- Distributed Systems: Hands-on experience scaling training/inference across 64+ GPUs (NCCL, MPI, Ray, Kubernetes/Slurm).
- LLM Lifecycle Mastery: End-to-end ownership of at least one production LLM (pre-training, SFT, RLHF/DPO, RAG, quantization, serving).
- Software Engineering Craft: Production-grade Python (type hints, async, profiling), Rust/C++ a plus; strong GitOps/Terraform habits.
- Research Fluency: Ability to read/reproduce papers (FlashAttention, Mixture of Experts, Speculative Decoding) and adapt novel techniques to production constraints.
- Communication: Ability to articulate complex trade-offs (latency vs. quality, cost vs. accuracy) to non-technical stakeholders.